MODEL PRICING

명확한 모델 가격과 포인트

모델 가격, 커넥터 연동 및 사용 가능 조건을 한 페이지에서 확인합니다. 실제 사용 가능 여부는 API 키 권한과 서버 경로에 따라 달라집니다.

페이지 범위

모델 가격, 커넥터 연동 및 사용 가능 조건을 한 페이지에서 확인합니다. 실제 사용 가능 여부는 API 키 권한과 서버 경로에 따라 달라집니다.

이 페이지는 해외 방문자를 위한 개요입니다. 기술 및 계약의 공식 기준은 중국어 버전입니다.

공개 카탈로그613개 모델 항목(후보 포함)

HCD Codex 커넥터91개 모델 수록

이미지 스튜디오25개 모델 수록

오디오 스튜디오17개 모델 수록

실시간 음성 스튜디오7개 모델 수록

비디오 스튜디오32개 모델 수록

문서 스튜디오6개 모델 수록

지식 스튜디오26개 모델 수록

3D 스튜디오3개 모델 수록

텍스트·코드 스튜디오2개 모델 수록

각 진입점의 공개된 모델 목록을 기준으로 스튜디오 내 중복 모델은 한 번만 집계합니다. 같은 모델이 여러 스튜디오에 포함될 수 있습니다. 실제 이용과 기능은 API 키 권한, 서버 경로 및 모델별 설명에 따라 달라집니다.

Completed capability declarations: 44. Each covers its stated deployment scope; model notes retain limits and pending items.

Token 요금은 백만 Token당 포인트입니다. 다른 기능은 표시된 단위로 계산합니다. 문맥 및 처리 단계는 각 모델의 요금을 따릅니다. 계정은 실제 사용량으로 계산하며 Key 계수는 Key 포인트에만 적용됩니다. 개방된 서비스는 실제 사용량에 따라 포인트를 차감합니다. 온라인 충전과 청구서 발행은 아직 제공하지 않습니다.

모델포인트 사용량

독립 모델을 우선 표시합니다. 같은 플랫폼·세대·기능 계열에서 사용 가능한 진입점이 하나뿐일 때만 스냅샷이나 별칭을 포함합니다. 모든 유형에서 다른 버전도 검색할 수 있으며 코드, 가격 및 사용 상태를 각각 유지합니다. 자료 대기나 유형 관계 미확정은 이용 불가를 의미하지 않습니다. 음색과 서비스 항목은 새 기반 모델이 아닙니다. 기존 API 경로와 인증 범위는 바뀌지 않습니다.

모델 유형표시 / 전체: 482 / 613

Claude Haiku 5.5claude-haiku-5-5독립 모델Anthropic · 정식2026-10-07전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능—가성비—

미평가

입력 사용량 · 백만 Token당200,000 포인트캐시 적중 사용량 · 백만 Token당20,000 포인트캐시 쓰기 사용량 · 5m · 백만 Token당250,000 포인트캐시 쓰기 사용량 · 1h · 백만 Token당400,000 포인트출력 사용량 · 백만 Token당1,000,000 포인트웹 검색 · 검색 단위 1,000개당20,000,000 포인트

Claude Haiku 5.5-100claude-haiku-5-5-100독립 모델Anthropic · 정식2026-10-07고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated74.7가성비 · Estimated82.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

미평가 · Confidence 0%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Gemini Nano Banana 2.1-100gemini-nano-banana-2.1독립 모델Google · 정식2026-10-06전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능92.5가성비 · Estimated66.5

잠정 평가

미평가 · Confidence 0%

입력 사용량 · 백만 Token당3,000,000 포인트입력 사용량 · 이미지 · 백만 Token당3,000,000 포인트캐시 적중 사용량 · 백만 Token당300,000 포인트출력 사용량 · 백만 Token당15,000,000 포인트캐시 적중 사용량 · 이미지 · 백만 Token당300,000 포인트출력 사용량 · 이미지 · 백만 Token당60,000,000 포인트

Qwen-Image-2.1-Proqwen-image-2.1-pro독립 모델Qwen · 정식2026-10-02전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated66.4가성비 · Estimated61.5

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen-image-2.0-pro-2026-06-22 · 66.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

출력 사용량 · 이미지 · 이미지당68,494 포인트

Grok 4.7-100grok-4.7-100독립 모델xAI · 미리보기2026-09-30전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능—가성비—

미평가

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당1,000,000 포인트출력 사용량 · 백만 Token당12,000,000 포인트

GPT-6.1 Solgpt-6.1-sol독립 모델OpenAI · 정식2026-09-29전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated82.2가성비 · Estimated69.2

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 76.4 (estimated) · 5/19 observed / reasoning 85.3 (estimated) · 3/13 observed / writing 84.8 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-6-astra · 1 benchmarksReference: gpt-6-luna · 1 benchmarksReference: gpt-6-sol · 1 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당200,000 포인트캐시 쓰기 사용량 · 백만 Token당5,000,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

Claude Sonnet 5.5claude-sonnet-5-5독립 모델Anthropic · 정식2026-09-28전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated74.4가성비 · Estimated63.8

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 84.5 (estimated) · 5/19 observed / reasoning 78.8 (estimated) · 3/13 observed / writing 59.8 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트캐시 쓰기 사용량 · 5m · 백만 Token당5,000,000 포인트캐시 쓰기 사용량 · 1h · 백만 Token당8,000,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트웹 검색 · 검색 단위 1,000개당20,000,000 포인트

Claude Sonnet 5.5-100claude-sonnet-5-5-100독립 모델Anthropic · 정식2026-09-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated74.4가성비 · Estimated67.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 84.5 (estimated) · 0/19 observed / reasoning 78.8 (estimated) · 0/13 observed / writing 59.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: claude-sonnet-5-5 · 12 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

决策模型（预览版）decision-model-preview독립 모델Qwen · 정식2026-09-24전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당0 포인트

Gemini 3.8 Live-100gemini-3.8-live독립 모델Google · 정식2026-09-24전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated76.9가성비 · Estimated68.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 66.9 (estimated) · 0/19 observed / reasoning 76.8 (estimated) · 0/13 observed / writing 87.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gemini-3.8-flash-high · 7 benchmarksReference: gemini-4-argon-high · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,500,000 포인트비디오 입력 · 백만 Token당2,000,000 포인트입력 사용량 · 오디오 · 백만 Token당6,000,000 포인트출력 사용량 · 백만 Token당9,000,000 포인트출력 사용량 · 오디오 · 백만 Token당24,000,000 포인트출력 사용량 · 비디오 생성 · 백만 Token당2,000,000 포인트

Claude Opus 5.5claude-opus-5-5독립 모델Anthropic · 정식2026-09-22전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated86.9가성비 · Estimated69.6

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 91.4 (estimated) · 5/19 observed / reasoning 94.0 (estimated) · 4/13 observed / writing 75.4 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당8,000,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트캐시 쓰기 사용량 · 5m · 백만 Token당10,000,000 포인트캐시 쓰기 사용량 · 1h · 백만 Token당16,000,000 포인트출력 사용량 · 백만 Token당40,000,000 포인트웹 검색 · 검색 단위 1,000개당20,000,000 포인트

Claude Opus 5.5-100claude-opus-5-5-100독립 모델Anthropic · 정식2026-09-22고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated86.9가성비 · Estimated75.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 91.4 (estimated) · 0/19 observed / reasoning 94.0 (estimated) · 0/13 observed / writing 75.4 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: claude-opus-5-5 · 13 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-6 Lunagpt-6-luna독립 모델OpenAI · 정식2026-09-22전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated59.4가성비 · Estimated65.7

Version comparison review13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 63.8 (estimated) · 5/19 observed / reasoning 64.0 (estimated) · 4/13 observed / writing 50.3 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReview: below gpt-5.6-luna by 4.3 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당200,000 포인트캐시 적중 사용량 · 백만 Token당20,000 포인트캐시 쓰기 사용량 · 백만 Token당250,000 포인트출력 사용량 · 백만 Token당1,000,000 포인트

GPT-6 Solgpt-6-sol독립 모델OpenAI · 정식2026-09-22전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated74.3가성비 · Estimated63.7

Version comparison review13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 74.6 (estimated) · 5/19 observed / reasoning 78.4 (estimated) · 4/13 observed / writing 69.9 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReview: below gpt-5.6-sol by 8.8 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트캐시 쓰기 사용량 · 백만 Token당5,000,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

Grok 4.7grok-4.7독립 모델xAI · 정식2026-09-21전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능—가성비—

미평가

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당1,000,000 포인트출력 사용량 · 백만 Token당12,000,000 포인트

Qwen-Audio-3.1-Realtime-Plusqwen-audio-3.1-realtime-plus독립 모델Qwen · 정식2026-09-20고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-realtime-preview-2024-12-17 · 45.3Reference: gpt-realtime-2025-08-28 · 59.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.openai-realtime-public · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.8-Livetranslate-Flash-Realtimeqwen3.8-livetranslate-flash-realtime독립 모델Qwen · 정식2026-09-18고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-MT-Uniqwen-mt-uni독립 모델Qwen · 정식2026-09-16고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated49.4가성비 · Estimated49.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 45.5 (estimated) · 0/19 observed / reasoning 54.8 (estimated) · 0/13 observed / writing 47.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-27b · 8 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen-glm-4.6 · 5 benchmarksReference: qwen-14b-chat · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-kimi/kimi-k2.8-previewqwen-kimi-kimi-k2.8-preview독립 모델Moonshot AI · 정식2026-09-11최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated29.9

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 0%coding 36.2 (estimated) · 0/19 observed / reasoning 50.2 (estimated) · 0/13 observed / writing 36.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: · 41.0Review: below qwen-minimax-speech-2.8-hd by 3.1 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,780,822 포인트캐시 적중 사용량 · 백만 Token당465,753 포인트출력 사용량 · 백만 Token당7,397,260 포인트

DeepSeek V4.1 Flashdeepseek-flash독립 모델DeepSeek · 정식2026-09-10전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated78.1가성비 · Estimated77.6

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 81.2 (estimated) · 4/19 observed / reasoning 82.4 (estimated) · 4/13 observed / writing 70.8 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksEstimated value uses this exact model's reviewed off-peak public point rates and the shared text workload. Weekday peak rates double; public holidays use off-peak rates. It is not a measured charge.arena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

입력 사용량 · 백만 Token당300,000 포인트캐시 적중 사용량 · 백만 Token당6,000 포인트출력 사용량 · 백만 Token당1,200,000 포인트

doubao-deepseek-v4-1-flashdoubao-deepseek-v4-1-flash-260910고정 스냅샷 / 버전 · 유일한 진입점DeepSeek · 정식2026-09-10전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated62.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당273,973 포인트캐시 적중 사용량 · 백만 Token당5,479 포인트출력 사용량 · 백만 Token당1,095,890 포인트

qwen-DeepSeek V4.1 Flashqwen-deepseek-v4.1-flash독립 모델DeepSeek · 정식2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.4가성비 · Estimated66.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 61.6 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 8 benchmarksReference: qwen3.5-flash · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당273,973 포인트캐시 적중 사용량 · 백만 Token당27,397 포인트출력 사용량 · 백만 Token당1,095,890 포인트

qwen-vanchin/deepseek-v4.1-flashqwen-vanchin-deepseek-v4.1-flash독립 모델DeepSeek · 정식2026-09-10고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated61.4가성비 · Estimated58.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 61.6 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 8 benchmarksReference: qwen3.5-flash · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT Image 2.5 Flaregpt-image-2.5-flare독립 모델OpenAI · 정식2026-09-08사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능98.5가성비70.3

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent arena · Independent

입력 사용량 · 백만 Token당10,000,000 포인트캐시 적중 사용량 · 백만 Token당2,500,000 포인트입력 사용량 · 이미지 · 백만 Token당16,000,000 포인트캐시 적중 사용량 · 이미지 · 백만 Token당4,000,000 포인트출력 사용량 · 이미지 · 백만 Token당60,000,000 포인트

GPT Image 2.5 Sunburstgpt-image-2.5-sunburst독립 모델OpenAI · 정식2026-09-08사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능100.0가성비71.4

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent arena · Independent

GPT-6 Astragpt-6-astra독립 모델OpenAI · 정식2026-09-03전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated85.4가성비 · Estimated64.8

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 79.9 (estimated) · 5/19 observed / reasoning 92.0 (estimated) · 4/13 observed / writing 84.3 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당20,000,000 포인트캐시 적중 사용량 · 백만 Token당2,000,000 포인트캐시 쓰기 사용량 · 백만 Token당25,000,000 포인트출력 사용량 · 백만 Token당100,000,000 포인트

Gemini 3.8 Flash-100gemini-3.8-flash-100독립 모델Google · 정식2026-09-02전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능—가성비—

미평가

입력 사용량 · 백만 Token당3,000,000 포인트입력 사용량 · 이미지 · 백만 Token당3,000,000 포인트입력 사용량 · 오디오 · 백만 Token당3,000,000 포인트비디오 입력 · 백만 Token당3,000,000 포인트캐시 적중 사용량 · 백만 Token당300,000 포인트출력 사용량 · 백만 Token당15,000,000 포인트캐시 쓰기 사용량 · 백만 Token당3,000,000 포인트명시적 캐시 저장 · 백만 Token·시간당2,000,000 포인트

Claude Fable 5.1claude-fable-5-1독립 모델Anthropic · 정식2026-09-01전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated89.2가성비 · Estimated67.4

Shared exact-model performance from claude-fable-5-1; coverage refers to public model benchmarks, not tests of every platform deployment.13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 90.5 (estimated) · 5/19 observed / reasoning 93.8 (estimated) · 4/13 observed / writing 83.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당20,000,000 포인트캐시 적중 사용량 · 백만 Token당500,000 포인트캐시 쓰기 사용량 · 5m · 백만 Token당25,000,000 포인트캐시 쓰기 사용량 · 1h · 백만 Token당40,000,000 포인트출력 사용량 · 백만 Token당100,000,000 포인트웹 검색 · 검색 단위 1,000개당20,000,000 포인트

Claude Fable 5.1-100claude-fable-5-1-100독립 모델Anthropic · 정식2026-09-01고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated89.2가성비—

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 90.5 (estimated) · 5/19 observed / reasoning 93.8 (estimated) · 4/13 observed / writing 83.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Doubao-Seed-2.1-Litedoubao-seed-2-1-lite-260915고정 스냅샷 / 버전 · 유일한 진입점Doubao · 정식2026-09 (공식 월 정보만 제공)최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated63.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 0%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: · 56.6microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당43,836 포인트출력 사용량 · 백만 Token당739,726 포인트

doubao-seedream-5-0-flashdoubao-seedream-5-0-flash-260915고정 스냅샷 / 버전 · 유일한 진입점Doubao · 정식2026-09 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated65.8가성비 · Estimated61.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: seedream-5.0-lite · 50.2Reference: seedream-5.0-pro · 85.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

千问3.7-通用文本向量-flashqwen3.7-text-embedding-flash독립 모델Alibaba Cloud · 정식2026-09-01최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.3가성비 · Estimated73.6

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: text-embedding-3-small · 62.3Reference: text-embedding-v3 · 63.4Reference: text-embedding-ada-002 · 61.0openai-embedding · Official aliyun-embedding · Official

임베딩 입력 · 백만 Token당34,247 포인트

千问3.7-通用文本重排序qwen3.7-text-rerank독립 모델Alibaba Cloud · 정식2026-09-01전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated72.5가성비 · Estimated80.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3-Reranker-4B · 73.9Reference: Qwen3-Reranker-8B · 74.2Reference: Qwen3-Reranker-0.6B · 68.8qwen-rerank-public · Official

입력 사용량 · 백만 Token당136,987 포인트

Qwen3.8-Omni-Flashqwen3.8-omni-flash독립 모델Qwen · 정식2026-09 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated61.4가성비 · Estimated58.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 61.6 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 8 benchmarksReference: qwen3.5-flash · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.8-Omni-Flash-Realtimeqwen3.8-omni-flash-realtime독립 모델Qwen · 정식2026-09 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-MT-Image-2.0qwen-mt-image-2.0독립 모델Qwen · 정식2026-08-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated73.8가성비 · Estimated66.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen-image-2.1 · 73.8Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Cohere Parse v5cohere-parse-v5독립 모델Cohere · 미리보기2026-08-27전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능79.2가성비85.4

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 18%cohere-parsebench · Official

문서 처리 · 1,000페이지당3,000,000 포인트

Gemini Omni 1.1 Flash Preview-100gemini-omni-1.1-flash-preview독립 모델Google · 정식2026-08-27고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated49.9가성비 · Estimated49.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 46.2 (estimated) · 0/19 observed / reasoning 44.4 (estimated) · 0/13 observed / writing 59.0 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gemini-3.5-flash-high · 7 benchmarksReference: gemini-1.5-flash-001 · 4 benchmarksReference: gemini-1.5-flash-002 · 4 benchmarksReference: gemini-1.5-flash-8b-001 · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ZHIPU/GLM-5.3-Flashqwen-zhipu-glm-5.3-flash독립 모델Zhipu AI · 정식2026-08-26최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated48.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 0%coding 36.2 (estimated) · 0/19 observed / reasoning 50.2 (estimated) · 0/13 observed / writing 36.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: · 41.0microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당63,014 포인트출력 사용량 · 백만 Token당767,123 포인트

Qwen3.8-Flashqwen3.8-flash독립 모델Qwen · 정식2026-08-25최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.4가성비 · Estimated67.5

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당27,397 포인트출력 사용량 · 백만 Token당739,726 포인트

qwen-Kimi-K3qwen-kimi-k3독립 모델Moonshot · 정식2026-08-19최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated39.7가성비 · Estimated38.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 35.6 (estimated) · 0/19 observed / reasoning 49.2 (estimated) · 0/13 observed / writing 34.4 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3-235b-a22b · 4 benchmarksReference: qwen3-30b-a3b · 4 benchmarksReference: qwen3-32b · 4 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당5,479,452 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당27,397,260 포인트

Qwen3.8-27Bqwen3.8-27b독립 모델Qwen · 정식2026-08-17최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated65.0가성비 · Estimated64.2

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 64.4 (estimated) · 5/19 observed / reasoning 68.3 (estimated) · 4/13 observed / writing 62.4 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당821,918 포인트캐시 적중 사용량 · 백만 Token당164,384 포인트출력 사용량 · 백만 Token당3,287,671 포인트

qwen-GLM-5.3qwen-glm-5.3독립 모델Zhipu AI · 정식2026-08-14최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.1가성비 · Estimated57.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 55.8 (estimated) · 0/19 observed / reasoning 64.7 (estimated) · 0/13 observed / writing 62.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-27b · 8 benchmarksReference: qwen-glm-4.6 · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

qwen-ZHIPU/GLM-5.3qwen-zhipu-glm-5.3독립 모델Zhipu AI · 정식2026-08-14최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated28.7

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

DeepSeek V4 Prodeepseek-v4-pro독립 모델DeepSeek · 정식2026-08-13전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated64.4가성비 · Estimated62.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 53.8 (estimated) · 0/19 observed / reasoning 74.3 (estimated) · 0/13 observed / writing 65.0 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: deepseek-v4-pro · 7 benchmarksReference: deepseek-v4-pro-0813 · 7 benchmarksReference: deepseek-v4-pro · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,320,000 포인트캐시 적중 사용량 · 백만 Token당44,000 포인트출력 사용량 · 백만 Token당3,960,000 포인트

GLM-5.3glm-5.3독립 모델Zhipu · 정식2026-08-13최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated71.7가성비 · Estimated65.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 73.9 (estimated) · 0/19 observed / reasoning 71.1 (estimated) · 0/13 observed / writing 70.2 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: glm-5.3 · 7 benchmarksReference: glm-5.2 · 6 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

GLM-5.3-Flashglm-5.3-flash독립 모델Zhipu · 정식2026-08-13최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated65.1가성비 · Estimated70.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 69.5 (estimated) · 0/19 observed / reasoning 68.7 (estimated) · 0/13 observed / writing 57.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: glm-5.3-flash · 12 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당63,014 포인트출력 사용량 · 백만 Token당767,123 포인트

Grok 4.6grok-4.6독립 모델xAI · 미리보기2026-08-12최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated70.4가성비 · Estimated61.9

Shared exact-model performance from grok-4.6; coverage refers to public model benchmarks, not tests of every platform deployment.13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 65.3 (estimated) · 5/19 observed / reasoning 72.6 (estimated) · 4/13 observed / writing 73.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당8,000,000 포인트캐시 적중 사용량 · 백만 Token당2,000,000 포인트출력 사용량 · 백만 Token당24,000,000 포인트

Grok 4.6-100grok-4.6-100독립 모델xAI · 정식2026-08-12고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated70.4가성비—

Shared exact-model performance from grok-4.6; coverage refers to public model benchmarks, not tests of every platform deployment.13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 65.3 (estimated) · 5/19 observed / reasoning 72.6 (estimated) · 4/13 observed / writing 73.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

MAI Thinking 1mai-thinking-1독립 모델Microsoft AI · 미리보기2026-08-12전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated51.7

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 18%coding 56.8 (estimated) · 4/19 observed / reasoning 60.2 (estimated) · 5/13 observed / writing 52.8 (estimated) · 3/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksReference: smaug-flash · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트출력 사용량 · 백만 Token당16,000,000 포인트

Qwen3.8-2.4T-A95Bqwen3.8-2.4t-a95b독립 모델Qwen · 정식2026-08-12최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated65.0가성비 · Estimated58.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 64.4 (estimated) · 0/19 observed / reasoning 68.3 (estimated) · 0/13 observed / writing 62.4 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen3.8-27b · 13 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당3,287,671 포인트캐시 적중 사용량 · 백만 Token당410,959 포인트출력 사용량 · 백만 Token당9,863,014 포인트

万相3.0-视频生成wan3.0-video독립 모델Alibaba Cloud · 정식2026-08-06최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated92.6가성비 · Estimated72.9

Estimate basis

비디오 생성 480P · 초당57,535 포인트비디오 생성 720P · 초당115,069 포인트비디오 생성 1080P · 초당230,137 포인트

Qwen3.8-Maxqwen3.8-max독립 모델Qwen · 정식2026-08-02최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated76.0가성비 · Estimated66.6

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 70.3 (estimated) · 5/19 observed / reasoning 82.3 (estimated) · 4/13 observed / writing 75.4 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당3,287,671 포인트캐시 적중 사용량 · 백만 Token당410,959 포인트출력 사용량 · 백만 Token당9,863,014 포인트

Gemini 3.5 Transcribe Live Preview-100gemini-3.5-transcribe-live-preview독립 모델Google · 정식2026-08 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Gemini 3.5 Transcribe Preview-100gemini-3.5-transcribe-preview독립 모델Google · 정식2026-08 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

doubao-DeepSeek V4 Flash GA (2026-07-31)doubao-deepseek-v4-flash-ga-260731고정 스냅샷 / 버전 · 유일한 진입점DeepSeek · 정식2026-07-31전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated58.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당821,918 포인트캐시 적중 사용량 · 백만 Token당27,397 포인트출력 사용량 · 백만 Token당2,465,753 포인트

qwen-MiniMax/MiniMax-H3qwen-minimax-minimax-h3독립 모델mini-max · 정식2026-07-31고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.0-ASR-Flashqwen-audio-3.0-asr-flash독립 모델Alibaba Cloud · 정식2026-07-30사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.4가성비 · Estimated94.0

Estimate basis

음성 전사 · 초당61 포인트

qwen-audio-3.0-asr-flash-filetransqwen-audio-3.0-asr-flash-filetrans독립 모델Qwen · 정식2026-07-30고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.0-ASR-Flash-Streamingqwen-audio-3.0-asr-flash-streaming독립 모델Qwen · 정식2026-07-30고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT Live Transcribegpt-live-transcribe독립 모델OpenAI · 정식2026-07-28전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.4가성비 · Estimated64.0

Estimate basis

음성 전사 · 시간당2,040,000 포인트

GPT Transcribegpt-transcribe독립 모델OpenAI · 정식2026-07-28사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.4가성비 · Estimated81.7

Estimate basis

음성 전사 · 시간당540,000 포인트

qwen-PixVerse-V6-r2v-omniqwen-pixverse-pixverse-v6-r2v-omni독립 모델pixverse · 정식2026-07-26고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated61.1가성비 · Estimated57.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Claude Opus 5claude-opus-5독립 모델Anthropic · 정식2026-07-24최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated81.9가성비 · Estimated65.2

Shared exact-model performance from claude-opus-5; coverage refers to public model benchmarks, not tests of every platform deployment.13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 85.2 (estimated) · 5/19 observed / reasoning 86.3 (estimated) · 4/13 observed / writing 74.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당10,000,000 포인트캐시 적중 사용량 · 백만 Token당1,000,000 포인트캐시 쓰기 사용량 · 5m · 백만 Token당12,500,000 포인트캐시 쓰기 사용량 · 1h · 백만 Token당20,000,000 포인트출력 사용량 · 백만 Token당50,000,000 포인트

Claude Opus 5-100claude-opus-5-100독립 모델Anthropic · 정식2026-07-24고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated81.9가성비—

Shared exact-model performance from claude-opus-5; coverage refers to public model benchmarks, not tests of every platform deployment.Version comparison review13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 85.2 (estimated) · 5/19 observed / reasoning 86.3 (estimated) · 4/13 observed / writing 74.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.Review: below claude-opus-5-5-100 by 5.0 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Gemini 3.5 Flash-Lite-100gemini-3.5-flash-lite-100독립 모델Google · 정식2026-07-21최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated52.3가성비 · Estimated55.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 47.1 (estimated) · 0/19 observed / reasoning 52.8 (estimated) · 0/13 observed / writing 56.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gemini-3.5-flash-lite-high · 7 benchmarksReference: gemini-3.5-flash-lite · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당600,000 포인트캐시 적중 사용량 · 백만 Token당60,000 포인트출력 사용량 · 백만 Token당5,000,000 포인트

Qwen3.7-Flashqwen3.7-flash독립 모델Qwen · 정식2026-07-21최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.4가성비 · Estimated73.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

Input 0 – 32,000 Token입력 사용량 · 백만 Token당54,795 포인트캐시 적중 사용량 · 백만 Token당10,959 포인트출력 사용량 · 백만 Token당219,178 포인트

Input 32,001 – 256,000 Token입력 사용량 · 백만 Token당164,384 포인트캐시 적중 사용량 · 백만 Token당32,877 포인트출력 사용량 · 백만 Token당657,534 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당328,767 포인트캐시 적중 사용량 · 백만 Token당65,753 포인트출력 사용량 · 백만 Token당1,315,068 포인트

Qwen-Image-3.0qwen-image-3.0독립 모델Alibaba Cloud · 정식2026-07-21사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated73.8가성비 · Estimated81.7

Estimate basis

이미지 입력 사용량 1K · 이미지당5,480 포인트이미지 입력 사용량 2K · 이미지당5,480 포인트이미지 출력 사용량 1K · 이미지당49,316 포인트이미지 출력 사용량 2K · 이미지당49,316 포인트

Kimi K3kimi-k3독립 모델Moonshot · 정식2026-07-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated80.1가성비 · Estimated66.4

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 81.7 (estimated) · 5/19 observed / reasoning 80.2 (estimated) · 4/13 observed / writing 78.3 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당5,479,452 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당27,397,260 포인트

qwen-kimi/kimi-k3qwen-kimi-kimi-k3독립 모델Moonshot AI · 정식2026-07-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated28.7

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당5,479,452 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당27,397,260 포인트

Qwen3.7-通用文本向量qwen3.7-text-embedding독립 모델Alibaba Cloud · 정식2026-07-15최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.3가성비 · Estimated53.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: text-embedding-3-small · 62.3Reference: text-embedding-v3 · 63.4Reference: text-embedding-ada-002 · 61.0openai-embedding · Official aliyun-embedding · Official

임베딩 입력 · 백만 Token당136,987 포인트

Qwen-Audio-3.0-Realtime-Flashqwen-audio-3.0-realtime-flash독립 모델Qwen · 정식2026-07-14고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.0-Realtime-Plusqwen-audio-3.0-realtime-plus독립 모델Qwen · 정식2026-07-14고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-5.6 Lunagpt-5.6-luna독립 모델OpenAI · 정식2026-07-09전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated63.7가성비 · Estimated65.5

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 64.4 (estimated) · 4/19 observed / reasoning 72.6 (estimated) · 4/13 observed / writing 54.0 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.6-luna-xhigh (codex-harness) · 1 benchmarksReference: gpt-5.6-sol-xhigh (codex-harness) · 1 benchmarksReference: gpt-5.6-terra-xhigh (codex-harness) · 1 benchmarksarena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

입력 사용량 · 백만 Token당400,000 포인트캐시 적중 사용량 · 백만 Token당40,000 포인트캐시 쓰기 사용량 · 백만 Token당500,000 포인트출력 사용량 · 백만 Token당2,400,000 포인트

GPT-5.6 Solgpt-5.6-sol독립 모델OpenAI · 정식2026-07-09전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated83.1가성비 · Estimated67.0

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 77.4 (estimated) · 4/19 observed / reasoning 89.4 (estimated) · 4/13 observed / writing 82.6 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.6-luna-xhigh (codex-harness) · 1 benchmarksReference: gpt-5.6-sol-xhigh (codex-harness) · 1 benchmarksReference: gpt-5.6-terra-xhigh (codex-harness) · 1 benchmarksarena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

입력 사용량 · 백만 Token당8,000,000 포인트캐시 적중 사용량 · 백만 Token당800,000 포인트캐시 쓰기 사용량 · 백만 Token당10,000,000 포인트출력 사용량 · 백만 Token당40,000,000 포인트

GPT-5.6 Terragpt-5.6-terra독립 모델OpenAI · 정식2026-07-09전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated73.1가성비 · Estimated62.5

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 65.0 (estimated) · 4/19 observed / reasoning 85.1 (estimated) · 4/13 observed / writing 69.3 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.6-luna-xhigh (codex-harness) · 1 benchmarksReference: gpt-5.6-sol-xhigh (codex-harness) · 1 benchmarksReference: gpt-5.6-terra-xhigh (codex-harness) · 1 benchmarksarena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트캐시 쓰기 사용량 · 백만 Token당5,000,000 포인트출력 사용량 · 백만 Token당24,000,000 포인트

gpt-live-1gpt-live-1독립 모델OpenAI · 정식2026-07-08고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated48.3가성비 · Estimated48.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 51.4 (estimated) · 0/19 observed / reasoning 52.2 (estimated) · 0/13 observed / writing 41.2 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: gpt-4-0125-preview · 10 benchmarksReference: gpt-4-1106-preview · 10 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4o · 7 benchmarksReference: gpt-5.2 · 7 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT Realtime 2.1gpt-realtime-2.1독립 모델OpenAI · 미리보기2026-07-06실시간 음성·텍스트 지원. 이미지 입력과 통화 중 도구 실행은 미지원입니다.최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated45.3가성비 · Estimated31.7

Estimate basis

입력 사용량 · 백만 Token당8,000,000 포인트캐시 적중 사용량 · 백만 Token당800,000 포인트출력 사용량 · 백만 Token당48,000,000 포인트입력 사용량 · 이미지 · 백만 Token당10,000,000 포인트캐시 적중 사용량 · 이미지 · 백만 Token당1,000,000 포인트입력 사용량 · 오디오 · 백만 Token당64,000,000 포인트캐시 적중 사용량 · 오디오 · 백만 Token당800,000 포인트출력 사용량 · 오디오 · 백만 Token당128,000,000 포인트

GPT Realtime 2.1 Minigpt-realtime-2.1-mini독립 모델OpenAI · 미리보기2026-07-06실시간 음성·텍스트 지원. 이미지 입력과 통화 중 도구 실행은 미지원입니다.최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated45.3가성비 · Estimated61.7

Estimate basis

입력 사용량 · 백만 Token당1,200,000 포인트캐시 적중 사용량 · 백만 Token당120,000 포인트출력 사용량 · 백만 Token당4,800,000 포인트입력 사용량 · 이미지 · 백만 Token당1,600,000 포인트캐시 적중 사용량 · 이미지 · 백만 Token당160,000 포인트입력 사용량 · 오디오 · 백만 Token당20,000,000 포인트캐시 적중 사용량 · 오디오 · 백만 Token당600,000 포인트출력 사용량 · 오디오 · 백만 Token당40,000,000 포인트

Qwen-Audio-3.0-TTS-Flashqwen-audio-3.0-tts-flash독립 모델Alibaba Cloud · 정식2026-07 (공식 월 정보만 제공)사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated74.4가성비 · Estimated74.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4qwen-tts-public · Official

오디오 · 10,000자당273,973 포인트

Qwen-Audio-3.0-TTS-Plusqwen-audio-3.0-tts-plus독립 모델Alibaba Cloud · 정식2026-07 (공식 월 정보만 제공)사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated74.4가성비 · Estimated63.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4qwen-tts-public · Official

오디오 · 10,000자당383,562 포인트

claude-sonnet-5claude-sonnet-5독립 모델Anthropic · 정식2026-06-30고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated70.2가성비 · Estimated64.1

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 76.0 (estimated) · 5/19 observed / reasoning 76.6 (estimated) · 4/13 observed / writing 58.1 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-sonnet-5-100claude-sonnet-5-100독립 모델Anthropic · 정식2026-06-30고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated70.2가성비 · Estimated64.1

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 76.0 (estimated) · 0/19 observed / reasoning 76.6 (estimated) · 0/13 observed / writing 58.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: claude-sonnet-5 · 13 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.Review: below claude-sonnet-5-5-100 by 4.2 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Doubao Seed 2.1 Turbo (2026-06-28)doubao-seed-2-1-turbo-260628고정 스냅샷 / 버전 · 유일한 진입점ByteDance · 정식2026-06-28최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated57.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당821,918 포인트캐시 적중 사용량 · 백만 Token당164,384 포인트출력 사용량 · 백만 Token당4,109,589 포인트

Mistral OCR 4mistral-ocr-4-0독립 모델Mistral AI · 미리보기2026-06-23전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능81.8가성비57.3

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 25%mistral-ocr · Official cohere-parsebench · Official

문서 처리 · 1,000페이지당8,000,000 포인트구조화 주석 포함 문서 처리 · 1,000페이지당10,000,000 포인트

HappyHorse-1.1-I2Vhappyhorse-1.1-i2v독립 모델Alibaba Cloud · 정식2026-06-22최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated80.9가성비 · Estimated56.6

Estimate basis

비디오 생성 480P · 초당123,288 포인트비디오 생성 720P · 초당246,576 포인트비디오 생성 1080P · 초당328,768 포인트

HappyHorse-1.1-R2Vhappyhorse-1.1-r2v독립 모델Alibaba Cloud · 정식2026-06-22최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated80.9가성비 · Estimated56.6

Estimate basis

비디오 생성 480P · 초당123,288 포인트비디오 생성 720P · 초당246,576 포인트비디오 생성 1080P · 초당328,768 포인트

HappyHorse-1.1-T2Vhappyhorse-1.1-t2v독립 모델Alibaba Cloud · 정식2026-06-22최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated80.9가성비 · Estimated56.6

Estimate basis

비디오 생성 480P · 초당123,288 포인트비디오 생성 720P · 초당246,576 포인트비디오 생성 1080P · 초당328,768 포인트

doubao-GLM 5.2 (2026-06-17)doubao-glm-5-2-260617고정 스냅샷 / 버전 · 유일한 진입점Zhipu · 정식2026-06-17최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated54.5

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

qwen-HappyOyster-1.0-Adventureqwen-happyoyster-1.0-adventure독립 모델happyoyster · 정식2026-06-17고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated43.2가성비 · Estimated45.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 37.0 (estimated) · 0/19 observed / reasoning 50.8 (estimated) · 0/13 observed / writing 41.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen-deepseek-r1 · 4 benchmarksReference: qwen-deepseek-r1-0528 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GLM-5.2glm-5.2독립 모델Zhipu · 정식2026-06-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated66.4가성비 · Estimated61.3

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 68.8 (estimated) · 5/19 observed / reasoning 70.9 (estimated) · 4/13 observed / writing 59.6 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

Qwen3.5-OCRqwen3.5-ocr독립 모델Alibaba Cloud · 정식2026-06-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated81.8가성비 · Estimated72.3

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mistral-ocr-4-0 · 81.8Reference: cohere-parse-v5 · 79.2Reference: mistral-document-ai-2512 · 96.1Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.mistral-ocr · Official cohere-parsebench · Official mistral-ocr · Official

입력 사용량 · 백만 Token당136,987 포인트출력 사용량 · 백만 Token당547,946 포인트

qwen-GLM-5.2qwen-glm-5.2독립 모델Zhipu · 정식2026-06-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.1가성비 · Estimated57.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 55.8 (estimated) · 0/19 observed / reasoning 64.7 (estimated) · 0/13 observed / writing 62.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-27b · 8 benchmarksReference: qwen-glm-4.6 · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

qwen-pixverse/pixverse-upscaleqwen-pixverse-pixverse-upscale독립 모델pixverse · 정식2026-06-16고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ZHIPU/GLM-5.2qwen-zhipu-glm-5.2독립 모델Zhipu AI · 정식2026-06-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated28.7

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

qwen-kimi-k2.7-codeqwen-kimi-k2.7-code독립 모델Moonshot · 정식2026-06-15최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated34.7가성비 · Estimated39.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 34.1 (estimated) · 0/19 observed / reasoning 41.4 (estimated) · 0/13 observed / writing 28.5 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen2.5-72b-instruct · 4 benchmarksReference: qwen2.5-coder-32b-instruct · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,780,822 포인트캐시 적중 사용량 · 백만 Token당356,164 포인트출력 사용량 · 백만 Token당7,397,260 포인트

Kimi K2.7 Codekimi-k2.7-code독립 모델Moonshot · 정식2026-06-12최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated46.3가성비 · Estimated47.8

8/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 47.8 (estimated) · 3/19 observed / reasoning 48.4 (estimated) · 3/13 observed / writing 42.8 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksarena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent

입력 사용량 · 백만 Token당1,780,822 포인트캐시 적중 사용량 · 백만 Token당356,164 포인트출력 사용량 · 백만 Token당7,397,260 포인트

qwen-kimi/kimi-k2.7-codeqwen-kimi-kimi-k2.7-code독립 모델Moonshot AI · 정식2026-06-12최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated29.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 0%coding 36.2 (estimated) · 0/19 observed / reasoning 50.2 (estimated) · 0/13 observed / writing 36.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: · 41.0microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,780,822 포인트캐시 적중 사용량 · 백만 Token당356,164 포인트출력 사용량 · 백만 Token당7,397,260 포인트

claude-fable-5claude-fable-5독립 모델Anthropic · 정식2026-06-09고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated87.7가성비 · Estimated76.4

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 85.6 (estimated) · 5/19 observed / reasoning 90.7 (estimated) · 4/13 observed / writing 86.7 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-fable-5-100claude-fable-5-100독립 모델Anthropic · 정식2026-06-09고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated87.7가성비 · Estimated76.4

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 85.6 (estimated) · 0/19 observed / reasoning 90.7 (estimated) · 0/13 observed / writing 86.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: claude-fable-5 · 13 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.Review: below claude-fable-5-1-100 by 1.5 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.7-Max-2026-06-08qwen3.7-max-2026-06-08고정 스냅샷 / 버전 · 유일한 진입점Qwen · 정식2026-06-09최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated64.5가성비 · Estimated58.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.2 (estimated) · 0/19 observed / reasoning 69.6 (estimated) · 0/13 observed / writing 73.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.7-max · 8 benchmarksReference: qwen3.7-max-preview · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당3,287,671 포인트캐시 적중 사용량 · 백만 Token당657,534 포인트출력 사용량 · 백만 Token당9,863,014 포인트

doubao-seedance-2-5-260628doubao-seedance-2-5-260628고정 스냅샷 / 버전 · 유일한 진입점ByteDance · 정식2026-06 (공식 월 정보만 제공)최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated90.4가성비 · Estimated63.3

Version comparison review

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: dreamina-seedance-2.5-720p · 90.4Review: below doubao-seedance-2-0-260128 by 1.1 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent arena · Independent

출력 사용량 · 비디오 생성 · 백만 Token당19,178,083 포인트

doubao-seed-evolvingdoubao-seed-evolving별칭 · 유일한 진입점Doubao · 정식2026-06 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

doubao-seedream-5-0-pro-260628doubao-seedream-5-0-pro-260628고정 스냅샷 / 버전 · 유일한 진입점ByteDance · 정식2026-06 (공식 월 정보만 제공)사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated85.2가성비 · Estimated69.2

Estimate basis

출력 사용량 · 이미지 · 이미지당82,192 포인트입력 사용량 · 이미지 · 이미지당5,480 포인트

Fun-ASR-Flash-2026-06-15fun-asr-flash-2026-06-15고정 스냅샷 / 버전 · 유일한 진입점Alibaba Cloud · 정식2026-06 (공식 월 정보만 제공)사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.4가성비 · Estimated94.0

Estimate basis

음성 전사 · 초당61 포인트

Qwen3.7-Plusqwen3.7-plus독립 모델Qwen · 정식2026-06-01최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.0가성비 · Estimated60.5

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.7 (estimated) · 2/19 observed / reasoning 62.0 (estimated) · 1/13 observed / writing 55.4 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-plus · 8 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 256,000 Token입력 사용량 · 백만 Token당438,356 포인트캐시 적중 사용량 · 백만 Token당87,671 포인트출력 사용량 · 백만 Token당1,753,425 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당1,315,068 포인트캐시 적중 사용량 · 백만 Token당263,014 포인트출력 사용량 · 백만 Token당5,260,274 포인트

qwen-MiniMax/MiniMax-M3qwen-minimax-minimax-m3독립 모델mini-max · 정식2026-06-01최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated46.0

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 36.2 (estimated) · 0/19 observed / reasoning 50.2 (estimated) · 0/13 observed / writing 36.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3-235b-a22b-thinking-2507 · 4 benchmarksReference: qwen3-30b-a3b · 4 benchmarksReference: qwen3-32b · 4 benchmarksReview: below qwen-minimax-speech-2.8-hd by 3.1 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,150,685 포인트캐시 적중 사용량 · 백만 Token당230,137 포인트출력 사용량 · 백만 Token당4,602,740 포인트

qwen-stepfun/step-3.7-flashqwen-stepfun-step-3.7-flash독립 모델stepfun · 정식2026-05-29최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.4가성비 · Estimated64.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당369,863 포인트캐시 적중 사용량 · 백만 Token당73,973 포인트출력 사용량 · 백만 Token당2,219,178 포인트

claude-opus-4-8claude-opus-4-8독립 모델Anthropic · 정식2026-05-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated74.3가성비 · Estimated67.0

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 72.3 (estimated) · 5/19 observed / reasoning 77.9 (estimated) · 4/13 observed / writing 72.6 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-opus-4-8-100claude-opus-4-8-100독립 모델Anthropic · 정식2026-05-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated74.3가성비 · Estimated67.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 72.3 (estimated) · 0/19 observed / reasoning 77.9 (estimated) · 0/13 observed / writing 72.6 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: claude-opus-4-8 · 13 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.7-Maxqwen3.7-max독립 모델Qwen · 정식2026-05-21최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated64.5가성비 · Estimated58.6

8/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.2 (estimated) · 3/19 observed / reasoning 69.6 (estimated) · 3/13 observed / writing 73.8 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.7-max-preview · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarkslivebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent

입력 사용량 · 백만 Token당3,287,671 포인트캐시 적중 사용량 · 백만 Token당657,534 포인트출력 사용량 · 백만 Token당9,863,014 포인트

Cohere Command A Pluscohere-command-a-plus-05-2026독립 모델Cohere · 미리보기2026-05-20최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated55.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,600,000 포인트출력 사용량 · 백만 Token당6,400,000 포인트

Qwen3.7-Max-Previewqwen3.7-max-preview독립 모델Qwen · 미리보기2026-05-20최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated64.5가성비 · Estimated58.6

4/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.2 (estimated) · 1/19 observed / reasoning 69.6 (estimated) · 1/13 observed / writing 73.8 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.7-max · 8 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당3,287,671 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당9,863,014 포인트

Qwen3.5-LiveTranslate-Flash-Realtimeqwen3.5-livetranslate-flash-realtime독립 모델Qwen · 정식2026-05-19고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Grok 4.3grok-4.3독립 모델xAI · 미리보기2026-05-12최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated44.8가성비 · Estimated46.5

Shared exact-model performance from grok-4.3; coverage refers to public model benchmarks, not tests of every platform deployment.12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 38.8 (estimated) · 4/19 observed / reasoning 48.3 (estimated) · 4/13 observed / writing 47.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트출력 사용량 · 백만 Token당5,000,000 포인트

Mistral Medium 3.5mistral-medium-3-5독립 모델Mistral AI · 미리보기2026-04-28전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated60.6가성비 · Estimated55.3

6/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.4 (estimated) · 3/19 observed / reasoning 70.3 (estimated) · 1/13 observed / writing 61.1 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksReference: deepseek-v4-pro-0813 · 2 benchmarksarena · Independent arena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official livebench · Independent

입력 사용량 · 백만 Token당3,000,000 포인트출력 사용량 · 백만 Token당15,000,000 포인트

HappyHorse-1.0-I2Vhappyhorse-1.0-i2v독립 모델Alibaba Cloud · 정식2026-04-27최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated80.9가성비 · Estimated56.6

Estimate basis

비디오 생성 720P · 초당246,576 포인트비디오 생성 1080P · 초당438,357 포인트

HappyHorse-1.0-R2Vhappyhorse-1.0-r2v독립 모델Alibaba Cloud · 정식2026-04-27최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated80.9가성비 · Estimated56.6

Estimate basis

비디오 생성 720P · 초당246,576 포인트비디오 생성 1080P · 초당438,357 포인트

HappyHorse-1.0-T2Vhappyhorse-1.0-t2v독립 모델Alibaba Cloud · 정식2026-04-27최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated80.9가성비 · Estimated56.6

Estimate basis

비디오 생성 720P · 초당246,576 포인트비디오 생성 1080P · 초당438,357 포인트

qwen-DeepSeek-V4-Proqwen-deepseek-v4-pro독립 모델DeepSeek · 정식2026-04-24전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated70.1가성비 · Estimated63.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 66.8 (estimated) · 0/19 observed / reasoning 74.5 (estimated) · 0/13 observed / writing 69.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen3.8-27b · 13 benchmarksReference: qwen3.8-max · 13 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 7 benchmarksReference: gpt-4.1-mini · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당3,287,671 포인트캐시 적중 사용량 · 백만 Token당273,973 포인트출력 사용량 · 백만 Token당6,575,342 포인트

GPT-5.5gpt-5.5독립 모델OpenAI · 정식2026-04-23최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated80.5가성비 · Estimated63.8

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 71.8 (estimated) · 4/19 observed / reasoning 88.6 (estimated) · 4/13 observed / writing 81.1 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.5 (codex-harness) · 1 benchmarksReference: gpt-5.5-high (codex-harness) · 1 benchmarksReference: gpt-5.5-xhigh (codex-harness) · 1 benchmarksarena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

입력 사용량 · 백만 Token당10,000,000 포인트캐시 적중 사용량 · 백만 Token당1,000,000 포인트출력 사용량 · 백만 Token당60,000,000 포인트

qwen-xiaomi/mimo-v2.5-proqwen-xiaomi-mimo-v2.5-pro독립 모델Xiaomi · 정식2026-04-23최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated30.5

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 0%coding 36.2 (estimated) · 0/19 observed / reasoning 50.2 (estimated) · 0/13 observed / writing 36.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: · 41.0microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,917,808 포인트캐시 적중 사용량 · 백만 Token당383,562 포인트출력 사용량 · 백만 Token당5,753,425 포인트

Qwen3.6-27Bqwen3.6-27b독립 모델Qwen · 정식2026-04-22최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated46.6가성비 · Estimated50.5

7/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 40.1 (estimated) · 2/19 observed / reasoning 56.9 (estimated) · 3/13 observed / writing 42.8 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.5-122b-a10b · 5 benchmarksReference: qwen3.5-27b · 5 benchmarksReference: qwen3.5-397b-a17b · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarkslivebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent

입력 사용량 · 백만 Token당821,918 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당4,931,507 포인트

GPT Image 2gpt-image-2독립 모델OpenAI · 정식2026-04-21사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능97.0가성비69.3

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent arena · Independent

qwen-Kimi-K2.6qwen-kimi-k2.6독립 모델Moonshot · 정식2026-04-21최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated34.9가성비 · Estimated39.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 34.1 (estimated) · 0/19 observed / reasoning 41.8 (estimated) · 0/13 observed / writing 28.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen2.5-72b-instruct · 4 benchmarksReference: qwen2.5-coder-32b-instruct · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,780,822 포인트캐시 적중 사용량 · 백만 Token당356,164 포인트출력 사용량 · 백만 Token당7,397,260 포인트

Kimi K2.6kimi-k2.6독립 모델Moonshot · 정식2026-04-20최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated77.4가성비 · Estimated69.6

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 77.8 (estimated) · 2/19 observed / reasoning 78.7 (estimated) · 1/13 observed / writing 75.7 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: kimi-k3 · 8 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,780,822 포인트캐시 적중 사용량 · 백만 Token당301,370 포인트출력 사용량 · 백만 Token당7,397,260 포인트

qwen-Kimi/Kimi K2.6qwen-kimi-kimi-k2.6독립 모델Moonshot AI · 정식2026-04-20최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated29.9

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당1,780,822 포인트캐시 적중 사용량 · 백만 Token당301,370 포인트출력 사용량 · 백만 Token당7,397,260 포인트

claude-opus-4-7claude-opus-4-7독립 모델Anthropic · 정식2026-04-16고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated74.1가성비 · Estimated66.9

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 73.6 (estimated) · 4/19 observed / reasoning 81.7 (estimated) · 4/13 observed / writing 67.0 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-opus-4-7-100claude-opus-4-7-100독립 모델Anthropic · 정식2026-04-16고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated74.1가성비 · Estimated66.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 73.6 (estimated) · 0/19 observed / reasoning 81.7 (estimated) · 0/13 observed / writing 67.0 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: claude-opus-4-7 · 12 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.6-35B-A3Bqwen3.6-35b-a3b독립 모델Qwen · 정식2026-04-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated45.4가성비 · Estimated51.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 39.0 (estimated) · 0/19 observed / reasoning 54.8 (estimated) · 0/13 observed / writing 42.3 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3.5-122b-a10b · 5 benchmarksReference: qwen3.5-35b-a3b · 5 benchmarksReference: qwen3.5-397b-a17b · 5 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당493,151 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당2,958,904 포인트

Qwen3.6-Flashqwen3.6-flash독립 모델Qwen · 정식2026-04-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated58.2가성비 · Estimated62.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 51.9 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 7 benchmarksReference: qwen3.5-flash · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 256,000 Token입력 사용량 · 백만 Token당328,767 포인트캐시 적중 사용량 · 백만 Token당65,753 포인트출력 사용량 · 백만 Token당1,972,603 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당1,315,068 포인트캐시 적중 사용량 · 백만 Token당263,014 포인트출력 사용량 · 백만 Token당7,890,411 포인트

qwen-GLM-5.1qwen-glm-5.1독립 모델Zhipu · 정식2026-04-14최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.1가성비 · Estimated58.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 55.8 (estimated) · 0/19 observed / reasoning 64.7 (estimated) · 0/13 observed / writing 62.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-27b · 8 benchmarksReference: qwen-glm-4.6 · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 32,000 Token입력 사용량 · 백만 Token당1,643,836 포인트캐시 적중 사용량 · 백만 Token당328,767 포인트출력 사용량 · 백만 Token당6,575,342 포인트

Input 32,001+ Token입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당438,356 포인트출력 사용량 · 백만 Token당7,671,233 포인트

qwen-ViduQ3-mix_reference2videoqwen-vidu-viduq3-mix_reference2video독립 모델vidu · 정식2026-04-13고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3_reference2videoqwen-vidu-viduq3_reference2video독립 모델vidu · 정식2026-04-13고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Turbo_reference2videoqwen-vidu-viduq3-turbo_reference2video독립 모델vidu · 정식2026-04-13고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-C1-it2vqwen-pixverse-pixverse-c1-it2v독립 모델pixverse · 정식2026-04-07고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-C1-kf2vqwen-pixverse-pixverse-c1-kf2v독립 모델pixverse · 정식2026-04-07고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-C1-r2vqwen-pixverse-pixverse-c1-r2v독립 모델pixverse · 정식2026-04-07고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-C1-t2vqwen-pixverse-pixverse-c1-t2v독립 모델pixverse · 정식2026-04-07고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ZHIPU/GLM-5.1qwen-zhipu-glm-5.1독립 모델Zhipu AI · 정식2026-04-07최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated28.7

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

Wan2.7-I2Vwan2.7-i2v독립 모델Alibaba Cloud · 정식2026-04-07최근 가격 확인
확인 대기 (UTC+8)

성능78.7가성비63.2

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

비디오 생성 720P · 초당164,384 포인트비디오 생성 1080P · 초당273,973 포인트

Wan2.7-R2Vwan2.7-r2v독립 모델Alibaba Cloud · 정식2026-04-07최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated69.2가성비 · Estimated56.5

Estimate basis

비디오 생성 720P · 초당164,384 포인트비디오 생성 1080P · 초당273,973 포인트

Wan2.7-T2Vwan2.7-t2v독립 모델Alibaba Cloud · 정식2026-04-07최근 가격 확인
확인 대기 (UTC+8)

성능59.6가성비49.8

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

비디오 생성 720P · 초당164,384 포인트비디오 생성 1080P · 초당273,973 포인트

qwen-Wan2.7-VideoEditwan2.7-videoedit독립 모델Wan · 정식2026-04-07고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated69.2가성비 · Estimated63.4

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.7-i2v · 78.7Reference: wan2.7-t2v · 59.6Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Grok 4.20 Non-Reasoninggrok-4-20-non-reasoning독립 모델xAI · 미리보기2026-04-06최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated54.7

Shared exact-model performance from grok-4-20-non-reasoning; coverage refers to public model benchmarks, not tests of every platform deployment.0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트출력 사용량 · 백만 Token당5,000,000 포인트

Grok 4.20 Reasoninggrok-4-20-reasoning독립 모델xAI · 미리보기2026-04-06최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated54.7

Shared exact-model performance from grok-4-20-reasoning; coverage refers to public model benchmarks, not tests of every platform deployment.0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트출력 사용량 · 백만 Token당5,000,000 포인트

Gemma 4 26B A4B It-100gemma-4-26b-a4b-it-maas-001독립 모델Google · 미리보기2026-04-02최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated62.5

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당300,000 포인트캐시 적중 사용량 · 백만 Token당30,000 포인트출력 사용량 · 백만 Token당1,200,000 포인트

Veo 3.1 Lite Generate 001-100veo-3.1-lite-generate-001독립 모델Google · 정식2026-04-02고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

doubao-seed3d-2-0-260328doubao-seed3d-2-0-260328고정 스냅샷 / 버전 · 유일한 진입점ByteDance · 정식2026-04 (공식 월 정보만 제공)최근 가격 확인
확인 대기 (UTC+8)

성능82.2가성비57.5

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 18%seed3d-public · Official

출력 사용량 · 3D · 백만 Token당21,917,809 포인트

Qwen3.6-Plusqwen3.6-plus독립 모델Qwen · 정식2026-04-01최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated54.5가성비 · Estimated57.7

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.4 (estimated) · 4/19 observed / reasoning 60.4 (estimated) · 4/13 observed / writing 52.8 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksarena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

Input 0 – 256,000 Token입력 사용량 · 백만 Token당547,945 포인트캐시 적중 사용량 · 백만 Token당109,589 포인트출력 사용량 · 백만 Token당3,287,671 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당438,356 포인트출력 사용량 · 백만 Token당13,150,685 포인트

Wan2.7-Imagewan2.7-image독립 모델Alibaba Cloud · 정식2026-04-01사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능41.7가성비46.3

Version comparison review

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%Review: below wan2.6-t2i by 8.3 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent arena · Independent

출력 사용량 · 이미지 · 이미지당54,795 포인트

Wan2.7-Image-Prowan2.7-image-pro독립 모델Alibaba Cloud · 정식2026-04-01사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능44.1가성비30.9

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent arena · Independent

출력 사용량 · 이미지 · 이미지당136,987 포인트

qwen-PixVerse-V6-it2vqwen-pixverse-pixverse-v6-it2v독립 모델pixverse · 정식2026-03-29고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-V6-kf2vqwen-pixverse-pixverse-v6-kf2v독립 모델pixverse · 정식2026-03-29고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-V6-t2vqwen-pixverse-pixverse-v6-t2v독립 모델pixverse · 정식2026-03-29고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GLM-5.1glm-5.1독립 모델Zhipu · 정식2026-03-27최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated65.1가성비 · Estimated61.4

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 66.7 (estimated) · 2/19 observed / reasoning 69.9 (estimated) · 1/13 observed / writing 58.6 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: glm-5.2 · 8 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 31,999 Token입력 사용량 · 백만 Token당1,643,836 포인트캐시 적중 사용량 · 백만 Token당356,164 포인트출력 사용량 · 백만 Token당6,575,342 포인트

Input 32,000+ Token입력 사용량 · 백만 Token당2,191,781 포인트캐시 적중 사용량 · 백만 Token당547,945 포인트출력 사용량 · 백만 Token당7,671,233 포인트

Lyria 3 Clip Preview-100lyria-3-clip-preview독립 모델Google · 정식2026-03-25고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Lyria 3 Pro Preview-100lyria-3-pro-preview독립 모델Google · 정식2026-03-25고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-MiniMax/MiniMax-M2.7qwen-minimax-minimax-m2.7독립 모델MiniMax · 정식2026-03-18최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated36.6

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당575,342 포인트캐시 적중 사용량 · 백만 Token당115,068 포인트출력 사용량 · 백만 Token당2,301,370 포인트

GPT-5.4 Minigpt-5.4-mini독립 모델OpenAI · 정식2026-03-17최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated47.9가성비 · Estimated48.9

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 44.3 (estimated) · 4/19 observed / reasoning 53.8 (estimated) · 4/13 observed / writing 45.6 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 7 benchmarksReference: gpt-5.4 · 1 benchmarksarena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

입력 사용량 · 백만 Token당1,500,000 포인트캐시 적중 사용량 · 백만 Token당150,000 포인트출력 사용량 · 백만 Token당9,000,000 포인트

GPT-5.4 Nanogpt-5.4-nano독립 모델OpenAI · 정식2026-03-17최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated48.6가성비 · Estimated54.8

11/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 44.5 (estimated) · 3/19 observed / reasoning 57.0 (estimated) · 4/13 observed / writing 44.3 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-nano-2025-04-14 · 7 benchmarksReference: gpt-5.4 · 2 benchmarksReference: gpt-5.4-high (codex-harness) · 1 benchmarksReference: gpt-5.4-mini · 1 benchmarksarena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent

입력 사용량 · 백만 Token당400,000 포인트캐시 적중 사용량 · 백만 Token당40,000 포인트출력 사용량 · 백만 Token당2,500,000 포인트

GLM-5-Turboglm-5-turbo독립 모델Zhipu · 정식2026-03-14최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated56.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 31,999 Token입력 사용량 · 백만 Token당1,369,863 포인트캐시 적중 사용량 · 백만 Token당328,767 포인트출력 사용량 · 백만 Token당6,027,397 포인트

Input 32,000+ Token입력 사용량 · 백만 Token당1,917,808 포인트캐시 적중 사용량 · 백만 Token당493,151 포인트출력 사용량 · 백만 Token당7,123,288 포인트

qwen-Tripo-P1.0qwen-tripo-tripo-p1.0독립 모델tripo · 정식2026-03-11고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated43.2가성비 · Estimated45.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Tripo-H3.1qwen-tripo-tripo-h3.1독립 모델tripo · 정식2026-03-06고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.9가성비 · Estimated46.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 37.4 (estimated) · 0/19 observed / reasoning 54.4 (estimated) · 0/13 observed / writing 42.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen-deepseek-v3.1 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-5.4gpt-5.4독립 모델OpenAI · 정식2026-03-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated73.0가성비 · Estimated61.5

13/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 58.7 (estimated) · 5/19 observed / reasoning 84.9 (estimated) · 4/13 observed / writing 75.5 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksarena · Independent arena · Independent livebench · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official

입력 사용량 · 백만 Token당5,000,000 포인트캐시 적중 사용량 · 백만 Token당500,000 포인트출력 사용량 · 백만 Token당30,000,000 포인트

GPT-5.4 Progpt-5.4-pro독립 모델OpenAI · 정식2026-03-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated58.8가성비 · Estimated41.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 49.8 (estimated) · 0/19 observed / reasoning 67.8 (estimated) · 0/13 observed / writing 58.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-5.4 · 13 benchmarksReference: gpt-5.4-mini · 12 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: gpt-5.4-nano · 11 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당60,000,000 포인트출력 사용량 · 백만 Token당360,000,000 포인트

Qwen3.5-Omni-Flashqwen3.5-omni-flash독립 모델Qwen · 정식2026-03 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated58.2가성비 · Estimated55.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 51.9 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 7 benchmarksReference: qwen3.5-flash · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.5-Omni-Flash-Realtimeqwen3.5-omni-flash-realtime독립 모델Qwen · 정식2026-03 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.5-Omni-Plusqwen3.5-omni-plus독립 모델Qwen · 정식2026-03 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated54.5가성비 · Estimated53.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.4 (estimated) · 0/19 observed / reasoning 60.4 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: qwen3.6-plus · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.5-Omni-Plus-Realtimeqwen3.5-omni-plus-realtime독립 모델Qwen · 정식2026-03 (공식 월 정보만 제공)고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-pixverse/pixverse-motioncontrolqwen-pixverse-pixverse-motioncontrol독립 모델pixverse · 정식2026-02-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-5.3 Codexgpt-5.3-codex독립 모델OpenAI · 정식2026-02-24최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated58.5가성비 · Estimated52.1

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.4 (estimated) · 0/19 observed / reasoning 63.1 (estimated) · 0/13 observed / writing 56.0 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-5.4 · 9 benchmarksReference: gpt-5.2 · 8 benchmarksReference: gpt-5.2-codex · 8 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReview: below gpt-5.2-codex by 3.5 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당3,500,000 포인트캐시 적중 사용량 · 백만 Token당350,000 포인트출력 사용량 · 백만 Token당28,000,000 포인트

qwen-MiniMax-M2.5qwen-minimax-m2.5독립 모델MiniMax · 정식2026-02-24최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated34.9가성비 · Estimated44.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당575,342 포인트캐시 적중 사용량 · 백만 Token당115,068 포인트출력 사용량 · 백만 Token당2,301,370 포인트

GPT Audio 1.5gpt-audio-1.5독립 모델OpenAI · 정식2026-02-23전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated52.6가성비 · Estimated41.6

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-realtime-preview-2024-12-17 · 45.3Reference: gpt-realtime-2025-08-28 · 59.9openai-realtime-public · Official

입력 사용량 · 백만 Token당5,000,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트입력 사용량 · 오디오 · 백만 Token당64,000,000 포인트출력 사용량 · 오디오 · 백만 Token당128,000,000 포인트

GPT Realtime 1.5gpt-realtime-1.5독립 모델OpenAI · 정식2026-02-23전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated45.3가성비 · Estimated31.7

Estimate basis

입력 사용량 · 백만 Token당8,000,000 포인트캐시 적중 사용량 · 백만 Token당800,000 포인트출력 사용량 · 백만 Token당32,000,000 포인트입력 사용량 · 이미지 · 백만 Token당10,000,000 포인트캐시 적중 사용량 · 이미지 · 백만 Token당1,000,000 포인트입력 사용량 · 오디오 · 백만 Token당64,000,000 포인트캐시 적중 사용량 · 오디오 · 백만 Token당800,000 포인트출력 사용량 · 오디오 · 백만 Token당128,000,000 포인트

Qwen3.5-122B-A10Bqwen3.5-122b-a10b독립 모델Qwen · 정식2026-02-23최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated46.4가성비 · Estimated55.2

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 39.9 (estimated) · 2/19 observed / reasoning 56.8 (estimated) · 1/13 observed / writing 42.4 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 128,000 Token입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당1,753,425 포인트

Input 128,001+ Token입력 사용량 · 백만 Token당547,945 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당4,383,562 포인트

Qwen3.5-27Bqwen3.5-27b독립 모델Qwen · 정식2026-02-23최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated45.7가성비 · Estimated55.9

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 39.4 (estimated) · 2/19 observed / reasoning 56.1 (estimated) · 1/13 observed / writing 41.7 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 128,000 Token입력 사용량 · 백만 Token당164,384 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당1,315,068 포인트

Input 128,001+ Token입력 사용량 · 백만 Token당493,151 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당3,945,205 포인트

Qwen3.5-35B-A3Bqwen3.5-35b-a3b독립 모델Qwen · 정식2026-02-23최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated42.9가성비 · Estimated55.6

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 36.8 (estimated) · 2/19 observed / reasoning 51.7 (estimated) · 1/13 observed / writing 40.2 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 128,000 Token입력 사용량 · 백만 Token당109,589 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당876,712 포인트

Input 128,001+ Token입력 사용량 · 백만 Token당438,356 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당3,506,849 포인트

Qwen3.5-Flashqwen3.5-flash독립 모델Qwen · 정식2026-02-23최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated58.2가성비 · Estimated68.6

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 51.9 (estimated) · 2/19 observed / reasoning 62.8 (estimated) · 1/13 observed / writing 59.9 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 7 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 128,000 Token입력 사용량 · 백만 Token당54,795 포인트캐시 적중 사용량 · 백만 Token당10,959 포인트출력 사용량 · 백만 Token당547,945 포인트

Input 128,001 – 256,000 Token입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당43,836 포인트출력 사용량 · 백만 Token당2,191,781 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당328,767 포인트캐시 적중 사용량 · 백만 Token당65,753 포인트출력 사용량 · 백만 Token당3,287,671 포인트

qwen-GLM-5qwen-glm-5독립 모델Zhipu · 정식2026-02-18최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.1가성비 · Estimated60.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 55.8 (estimated) · 0/19 observed / reasoning 64.7 (estimated) · 0/13 observed / writing 62.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-27b · 8 benchmarksReference: qwen-glm-4.6 · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 32,000 Token입력 사용량 · 백만 Token당1,095,890 포인트캐시 적중 사용량 · 백만 Token당219,178 포인트출력 사용량 · 백만 Token당4,931,507 포인트

Input 32,001+ Token입력 사용량 · 백만 Token당1,643,836 포인트캐시 적중 사용량 · 백만 Token당328,767 포인트출력 사용량 · 백만 Token당6,027,397 포인트

claude-sonnet-4-6claude-sonnet-4-6독립 모델Anthropic · 정식2026-02-17고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated65.2가성비 · Estimated60.6

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 64.0 (estimated) · 4/19 observed / reasoning 71.9 (estimated) · 4/13 observed / writing 59.7 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-sonnet-4-6-100claude-sonnet-4-6-100독립 모델Anthropic · 정식2026-02-17고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated65.2가성비 · Estimated60.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 64.0 (estimated) · 0/19 observed / reasoning 71.9 (estimated) · 0/13 observed / writing 59.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: claude-sonnet-4-6 · 12 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3.5-397B-A17Bqwen3.5-397b-a17b독립 모델Qwen · 정식2026-02-15최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated49.6가성비 · Estimated56.4

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 42.6 (estimated) · 2/19 observed / reasoning 59.5 (estimated) · 1/13 observed / writing 46.6 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 128,000 Token입력 사용량 · 백만 Token당328,767 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당1,972,603 포인트

Input 128,001+ Token입력 사용량 · 백만 Token당821,918 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당4,931,507 포인트

Qwen3.5-Plusqwen3.5-plus독립 모델Qwen · 정식2026-02-15최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated54.5가성비 · Estimated61.5

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.4 (estimated) · 0/19 observed / reasoning 60.4 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: qwen3.6-plus · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 128,000 Token입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당43,836 포인트출력 사용량 · 백만 Token당1,315,068 포인트

Input 128,001 – 256,000 Token입력 사용량 · 백만 Token당547,945 포인트캐시 적중 사용량 · 백만 Token당109,589 포인트출력 사용량 · 백만 Token당3,287,671 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당1,095,890 포인트캐시 적중 사용량 · 백만 Token당219,178 포인트출력 사용량 · 백만 Token당6,575,342 포인트

qwen-MiniMax/MiniMax-M2.5qwen-minimax-minimax-m2.5독립 모델MiniMax · 정식2026-02-12최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated36.6

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당575,342 포인트캐시 적중 사용량 · 백만 Token당57,534 포인트출력 사용량 · 백만 Token당2,301,370 포인트

qwen-PixVerse-V5.6-r2vqwen-pixverse-pixverse-v5.6-r2v독립 모델pixverse · 정식2026-02-12고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ZHIPU/GLM-5qwen-zhipu-glm-5독립 모델Zhipu AI · 정식2026-02-12최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated31.2

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당1,643,836 포인트캐시 적중 사용량 · 백만 Token당410,959 포인트출력 사용량 · 백만 Token당6,027,397 포인트

qwen-ViduQ3-Pro_start-end2videoqwen-vidu-viduq3-pro_start-end2video독립 모델vidu · 정식2026-02-11고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Turbo_img2videoqwen-vidu-viduq3-turbo_img2video독립 모델vidu · 정식2026-02-11고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Turbo_start-end2videoqwen-vidu-viduq3-turbo_start-end2video독립 모델vidu · 정식2026-02-11고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Turbo_text2videoqwen-vidu-viduq3-turbo_text2video독립 모델vidu · 정식2026-02-11고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GLM-5glm-5독립 모델Zhipu · 정식2026-02-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.3가성비 · Estimated60.9

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 63.4 (estimated) · 2/19 observed / reasoning 66.5 (estimated) · 1/13 observed / writing 56.9 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: glm-5.2 · 8 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 31,999 Token입력 사용량 · 백만 Token당1,095,890 포인트캐시 적중 사용량 · 백만 Token당273,973 포인트출력 사용량 · 백만 Token당4,931,507 포인트

Input 32,000+ Token입력 사용량 · 백만 Token당1,643,836 포인트캐시 적중 사용량 · 백만 Token당410,959 포인트출력 사용량 · 백만 Token당6,027,397 포인트

Qwen-Image-2.0qwen-image-2.0독립 모델Alibaba Cloud · 정식2026-02-10사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated73.8가성비 · Estimated68.7

Estimate basis

출력 사용량 · 이미지 · 이미지당54,795 포인트

claude-opus-4-6claude-opus-4-6독립 모델Anthropic · 정식2026-02-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated70.8가성비 · Estimated64.6

12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 66.4 (estimated) · 4/19 observed / reasoning 75.9 (estimated) · 4/13 observed / writing 70.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-opus-4-6-100claude-opus-4-6-100독립 모델Anthropic · 정식2026-02-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated70.8가성비 · Estimated64.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 66.4 (estimated) · 0/19 observed / reasoning 75.9 (estimated) · 0/13 observed / writing 70.2 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: claude-opus-4-6 · 12 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Kling Image 3.0qwen-kling-kling-v3-image-generation독립 모델kling · 정식2026-02-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated73.8가성비 · Estimated66.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen-image-2.1 · 73.8Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Kling Image 3.0 Omniqwen-kling-kling-v3-omni-image-generation독립 모델kling · 정식2026-02-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Kling Video 3.0 Omniqwen-kling-kling-v3-omni-video-generation독립 모델kling · 정식2026-02-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated41.7가성비 · Estimated44.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 36.7 (estimated) · 0/19 observed / reasoning 50.8 (estimated) · 0/13 observed / writing 37.6 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3-235b-a22b · 4 benchmarksReference: qwen3-30b-a3b · 4 benchmarksReference: qwen3-vl-235b-a22b-instruct · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Kling Video 3.0qwen-kling-kling-v3-video-generation독립 모델kling · 정식2026-02-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Kimi-K2.5qwen-kimi-k2.5독립 모델Moonshot · 정식2026-01-30최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated34.9가성비 · Estimated41.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당1,095,890 포인트캐시 적중 사용량 · 백만 Token당219,178 포인트출력 사용량 · 백만 Token당5,753,425 포인트

Fun-ASR-Flash-8k实时语音识别fun-asr-flash-8k-realtime독립 모델Qwen · 정식2026-01-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2-Pro_reference2videoqwen-vidu-viduq2-pro_reference2video독립 모델vidu · 정식2026-01-27고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Pro_img2videoqwen-vidu-viduq3-pro_img2video독립 모델vidu · 정식2026-01-27고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Pro_text2videoqwen-vidu-viduq3-pro_text2video독립 모델vidu · 정식2026-01-27고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3-VL-Plusqwen3-vl-plus독립 모델Qwen · 정식2026-01-26최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated42.7가성비 · Estimated51.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 45.3 (estimated) · 0/19 observed / reasoning 44.7 (estimated) · 0/13 observed / writing 38.0 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-plus · 8 benchmarksReference: qwen2.5-plus-1127 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 32,000 Token입력 사용량 · 백만 Token당273,973 포인트캐시 적중 사용량 · 백만 Token당54,795 포인트출력 사용량 · 백만 Token당2,739,726 포인트

Input 32,001 – 128,000 Token입력 사용량 · 백만 Token당410,959 포인트캐시 적중 사용량 · 백만 Token당82,192 포인트출력 사용량 · 백만 Token당4,109,589 포인트

Input 128,001+ Token입력 사용량 · 백만 Token당821,918 포인트캐시 적중 사용량 · 백만 Token당164,384 포인트출력 사용량 · 백만 Token당8,219,178 포인트

qwen-PixVerse-V5.6-it2vqwen-pixverse-pixverse-v5.6-it2v독립 모델pixverse · 정식2026-01-26고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-V5.6-kf2vqwen-pixverse-pixverse-v5.6-kf2v독립 모델pixverse · 정식2026-01-26고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-V5.6-t2vqwen-pixverse-pixverse-v5.6-t2v독립 모델pixverse · 정식2026-01-26고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Virtual Try On 001-100virtual-try-on-001독립 모델Google · 정식2026-01-23고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.5가성비 · Estimated46.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.7-image-pro · 44.1Reference: krea-2-medium · 45.0Reference: reve-v1.1 · 43.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-5.2 Codexgpt-5.2-codex독립 모델OpenAI · 정식2026-01-14최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.0가성비 · Estimated54.5

Version comparison review8/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 59.8 (estimated) · 3/19 observed / reasoning 68.2 (estimated) · 3/13 observed / writing 58.0 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.2 · 4 benchmarksReference: gpt-5.2-chat-latest-20260210 · 4 benchmarksReference: gpt-5.4 · 1 benchmarksReview: below gpt-5.1-codex by 1.6 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent

입력 사용량 · 백만 Token당3,500,000 포인트캐시 적중 사용량 · 백만 Token당350,000 포인트출력 사용량 · 백만 Token당28,000,000 포인트

qwen-MiniMax/MiniMax-M2.1qwen-minimax-minimax-m2.1독립 모델MiniMax · 정식2025-12-23최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated36.6

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당575,342 포인트캐시 적중 사용량 · 백만 Token당57,534 포인트출력 사용량 · 백만 Token당2,301,370 포인트

GLM-4.7glm-4.7독립 모델Zhipu · 정식2025-12-22최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated60.2가성비 · Estimated60.6

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 63.3 (estimated) · 2/19 observed / reasoning 63.5 (estimated) · 1/13 observed / writing 53.7 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: glm-5.2 · 8 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

Input 0 – 31,999 Token · Output 0 – 199 Token입력 사용량 · 백만 Token당547,945 포인트캐시 적중 사용량 · 백만 Token당109,589 포인트출력 사용량 · 백만 Token당2,191,781 포인트

Input 0 – 31,999 Token · Output 200+ Token입력 사용량 · 백만 Token당821,918 포인트캐시 적중 사용량 · 백만 Token당164,384 포인트출력 사용량 · 백만 Token당3,835,616 포인트

Input 32,000+ Token입력 사용량 · 백만 Token당1,095,890 포인트캐시 적중 사용량 · 백만 Token당219,178 포인트출력 사용량 · 백만 Token당4,383,562 포인트

GPT Image 1.5gpt-image-1.5독립 모델OpenAI · 정식2025-12-16사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능76.0가성비53.2

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent arena · Independent

입력 사용량 · 백만 Token당10,000,000 포인트캐시 적중 사용량 · 백만 Token당2,500,000 포인트입력 사용량 · 이미지 · 백만 Token당16,000,000 포인트캐시 적중 사용량 · 이미지 · 백만 Token당4,000,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트출력 사용량 · 이미지 · 백만 Token당64,000,000 포인트

Wan2.6-I2Vwan2.6-i2v독립 모델Alibaba Cloud · 정식2025-12-16최근 가격 확인
확인 대기 (UTC+8)

성능53.2가성비45.3

Version comparison review

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%Review: below wan2.5-i2v-preview by 2.1 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent

비디오 생성 720P · 초당164,384 포인트비디오 생성 1080P · 초당273,973 포인트

Wan2.6-Imagewan2.6-image독립 모델Alibaba Cloud · 정식2025-12-16사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능31.6가성비39.2

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

출력 사용량 · 이미지 · 이미지당54,795 포인트

qwen-Wan2.6-R2Vwan2.6-r2v독립 모델Wan · 정식2025-12-16고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated55.3가성비 · Estimated53.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.6-i2v · 53.2Reference: wan2.6-t2v · 57.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Wan2.6-T2Iwan2.6-t2i독립 모델Alibaba Cloud · 정식2025-12-16사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능50.0가성비52.1

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

출력 사용량 · 이미지 · 이미지당54,795 포인트

Wan2.6-T2Vwan2.6-t2v독립 모델Alibaba Cloud · 정식2025-12-16최근 가격 확인
확인 대기 (UTC+8)

성능57.4가성비48.3

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

비디오 생성 720P · 초당164,384 포인트비디오 생성 1080P · 초당273,973 포인트

Cohere Rerank v4.0 Procohere-rerank-v4.0-pro독립 모델Cohere · 정식2025-12-11전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated65.8가성비 · Estimated61.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3-Reranker-0.6B · 68.8Reference: gte-multilingual-reranker-base · 62.7Reference: BGE-reranker-v2-m3 · 57.7qwen-rerank-public · Official

재순위 · 검색 단위 1,000개당5,000,000 포인트

GPT-5.2gpt-5.2독립 모델OpenAI · 정식2025-12-11최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated61.7가성비 · Estimated54.3

Version comparison review12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 52.5 (estimated) · 4/19 observed / reasoning 73.8 (estimated) · 4/13 observed / writing 58.7 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.4 · 1 benchmarksReview: below gpt-5.1 by 2.1 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

입력 사용량 · 백만 Token당3,500,000 포인트캐시 적중 사용량 · 백만 Token당350,000 포인트출력 사용량 · 백만 Token당28,000,000 포인트

GPT-5.1 Codex Maxgpt-5.1-codex-max독립 모델OpenAI · 정식2025-12-04최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated72.5가성비 · Estimated63.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 69.3 (estimated) · 0/19 observed / reasoning 78.4 (estimated) · 0/13 observed / writing 69.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-6-sol-max · 12 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.4 · 1 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당250,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

Mistral Large 3mistral-large-3독립 모델Mistral AI · 정식2025-12-02최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated60.6가성비 · Estimated60.8

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 54.7 (estimated) · 2/19 observed / reasoning 66.7 (estimated) · 1/13 observed / writing 60.5 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksReference: deepseek-v4-pro-0813 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

입력 사용량 · 백만 Token당1,000,000 포인트출력 사용량 · 백만 Token당3,000,000 포인트

qwen-ViduQ2-Pro-fast_img2videoqwen-vidu-viduq2-pro-fast_img2video독립 모델vidu · 정식2025-12-02고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3-LiveTranslate-Flashqwen3-livetranslate-flash독립 모델Qwen · 정식2025-12-01고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated58.2가성비 · Estimated55.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 51.9 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 7 benchmarksReference: qwen3.5-flash · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3-Omni-Flashqwen3-omni-flash독립 모델Qwen · 정식2025-12-01고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated58.2가성비 · Estimated55.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 51.9 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 7 benchmarksReference: qwen3.5-flash · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-SiliconFlow DeepSeek-V3.2qwen-siliconflow-deepseek-v3.2독립 모델DeepSeek · 정식2025-12-01고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated45.8가성비 · Estimated47.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 38.1 (estimated) · 0/19 observed / reasoning 54.2 (estimated) · 0/13 observed / writing 45.0 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen-deepseek-v3.2-exp · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Vanchin/DeepSeek-V3.2-thinkqwen-vanchin-deepseek-v3.2-think독립 모델DeepSeek · 정식2025-12-01고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated45.8가성비 · Estimated47.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3-TTS-Flash-Realtimeqwen3-tts-flash-realtime독립 모델Qwen · 정식2025-11-27고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Z-Image-Turboz-image-turbo독립 모델Alibaba Cloud · 정식2025-11-26사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능28.8가성비50.2

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

출력 사용량 · 이미지 · 이미지당27,398 포인트

FLUX.2 [flex]flux.2-flex독립 모델Black Forest Labs · 정식2025-11-25사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능46.8가성비32.8

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent arena · Independent

입력 사용량 · 이미지 · 메가픽셀당100,000 포인트출력 사용량 · 이미지 · 메가픽셀당100,000 포인트

FLUX.2 Proflux.2-pro독립 모델Black Forest Labs · 정식2025-11-25전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능49.9가성비64.9

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent arena · Independent

첫 메가픽셀 · 메가픽셀당60,000 포인트추가 메가픽셀 · 메가픽셀당30,000 포인트

claude-opus-4-5claude-opus-4-5독립 모델Anthropic · 정식2025-11-24고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated67.9가성비—

Shared exact-model performance from claude-opus-4-5; coverage refers to public model benchmarks, not tests of every platform deployment.0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 63.9 (estimated) · 0/19 observed / reasoning 71.8 (estimated) · 0/13 observed / writing 67.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: claude-opus-4-6 · 7 benchmarksReference: claude-opus-4-5-20251101 · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-opus-4-5-100claude-opus-4-5-100독립 모델Anthropic · 정식2025-11-24고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated67.9가성비—

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Grok 4.1 Fast Non-Reasoninggrok-4-1-fast-non-reasoning독립 모델xAI · 정식2025-11-19최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated62.1

Shared exact-model performance from grok-4-1-fast-non-reasoning; coverage refers to public model benchmarks, not tests of every platform deployment.0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당400,000 포인트캐시 적중 사용량 · 백만 Token당100,000 포인트출력 사용량 · 백만 Token당1,000,000 포인트

Grok 4.1 Fast Reasoninggrok-4-1-fast-reasoning독립 모델xAI · 정식2025-11-19최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated59.5가성비 · Estimated64.1

Shared exact-model performance from grok-4-1-fast-reasoning; coverage refers to public model benchmarks, not tests of every platform deployment.5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 53.3 (estimated) · 2/19 observed / reasoning 65.8 (estimated) · 1/13 observed / writing 59.5 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksReference: deepseek-v4-pro-0813 · 2 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

입력 사용량 · 백만 Token당400,000 포인트캐시 적중 사용량 · 백만 Token당100,000 포인트출력 사용량 · 백만 Token당1,000,000 포인트

Gemini 3 Pro Preview-100gemini-3-pro-preview독립 모델Google · 정식2025-11-18고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated69.4가성비 · Estimated63.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 48.1 (estimated) · 0/19 observed / reasoning 77.5 (estimated) · 0/13 observed / writing 82.6 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gemini-3.1-pro-preview-high · 7 benchmarksReference: gemini-3-pro · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Veo 3.1 Fast Generate 001-100veo-3.1-fast-generate-001독립 모델Google · 정식2025-11-17고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Veo 3.1 Generate 001-100veo-3.1-generate-001독립 모델Google · 정식2025-11-17고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-5.1gpt-5.1독립 모델OpenAI · 정식2025-11-13최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated63.8가성비 · Estimated57.2

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 54.9 (estimated) · 2/19 observed / reasoning 73.2 (estimated) · 1/13 observed / writing 63.4 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.2 · 7 benchmarksReference: gpt-5.2-codex · 7 benchmarksReference: gpt-5.4 · 1 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당250,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

GPT-5.1 Codexgpt-5.1-codex독립 모델OpenAI · 정식2025-11-13최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated63.6가성비 · Estimated57.1

1/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 58.2 (estimated) · 1/19 observed / reasoning 70.3 (estimated) · 0/13 observed / writing 62.4 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.2 · 7 benchmarksReference: gpt-5.2-codex · 7 benchmarksReference: gpt-5.1 · 4 benchmarksarena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당250,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

GPT-5.1 Codex Minigpt-5.1-codex-mini독립 모델OpenAI · 정식2025-11-13최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated43.1가성비 · Estimated49.4

1/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 39.6 (estimated) · 1/19 observed / reasoning 48.9 (estimated) · 0/13 observed / writing 40.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 7 benchmarksReference: gpt-5.4-mini · 7 benchmarksReference: gpt-5-mini · 4 benchmarksReference: gpt-5.4 · 1 benchmarksarena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당500,000 포인트캐시 적중 사용량 · 백만 Token당50,000 포인트출력 사용량 · 백만 Token당4,000,000 포인트

Lyria 002-100lyria-002독립 모델Google · 정식2025-10-27고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2_reference2videoqwen-vidu-viduq2_reference2video독립 모델vidu · 정식2025-10-21고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2_text2videoqwen-vidu-viduq2_text2video독립 모델vidu · 정식2025-10-21고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Vanchin/DeepSeek-OCRqwen-vanchin-deepseek-ocr독립 모델DeepSeek · 정식2025-10-20고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated81.8가성비 · Estimated72.3

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mistral-ocr-4-0 · 81.8Reference: cohere-parse-v5 · 79.2Reference: mistral-document-ai-2512 · 96.1Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.mistral-ocr · Official cohere-parsebench · Official mistral-ocr · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-haiku-4-5claude-haiku-4-5독립 모델Anthropic · 정식2025-10-15고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated60.4가성비 · Estimated57.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.5 (estimated) · 0/19 observed / reasoning 63.8 (estimated) · 0/13 observed / writing 60.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: claude-haiku-4-5-20251001 · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

claude-haiku-4-5-100claude-haiku-4-5-100독립 모델Anthropic · 정식2025-10-15고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated60.4가성비 · Estimated57.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-5 Progpt-5-pro독립 모델OpenAI · 정식2025-10-06최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.5가성비 · Estimated41.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 55.9 (estimated) · 0/19 observed / reasoning 61.9 (estimated) · 0/13 observed / writing 51.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.2 · 7 benchmarksReference: gpt-5.2-codex · 7 benchmarksReference: gpt-5 · 5 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당20,000,000 포인트출력 사용량 · 백만 Token당240,000,000 포인트

GPT Audio Minigpt-audio-mini독립 모델OpenAI · 정식2025-10-06전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated52.6가성비 · Estimated66.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-realtime-preview-2024-12-17 · 45.3Reference: gpt-realtime-2025-08-28 · 59.9openai-realtime-public · Official

입력 사용량 · 백만 Token당1,200,000 포인트출력 사용량 · 백만 Token당4,800,000 포인트입력 사용량 · 오디오 · 백만 Token당20,000,000 포인트출력 사용량 · 오디오 · 백만 Token당40,000,000 포인트

GPT Image 1 Minigpt-image-1-mini독립 모델OpenAI · 정식2025-10-06사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능20.1가성비44.1

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent arena · Independent

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트입력 사용량 · 이미지 · 백만 Token당5,000,000 포인트캐시 적중 사용량 · 이미지 · 백만 Token당500,000 포인트출력 사용량 · 이미지 · 백만 Token당16,000,000 포인트

GPT Realtime Minigpt-realtime-mini독립 모델OpenAI · 정식2025-10-06전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated54.0가성비 · Estimated67.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-realtime-2025-08-28 · 59.9Reference: gpt-4o-realtime-preview-2024-12-17 · 45.3openai-realtime-public · Official

Gemini 2.5 Flash TTS-100gemini-2.5-flash-tts독립 모델Google · 정식2025-09-30고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated87.1가성비 · Estimated76.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Gemini-flash · 87.1Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-tts-public · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Gemini 2.5 Pro TTS-100gemini-2.5-pro-tts독립 모델Google · 정식2025-09-30고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated82.6가성비 · Estimated72.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Gemini-pro · 82.6Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-tts-public · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sora 2sora-2독립 모델OpenAI · 미리보기2025-09-30최근 가격 확인
확인 대기 (UTC+8)

성능66.0가성비50.4

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

비디오 생성 · 초당200,000 포인트

claude-sonnet-4-5claude-sonnet-4-5독립 모델Anthropic · 정식2025-09-29고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated61.6가성비 · Estimated58.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 58.8 (estimated) · 0/19 observed / reasoning 66.2 (estimated) · 0/13 observed / writing 59.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: claude-sonnet-4-6 · 7 benchmarksReference: claude-sonnet-4-5-20250929 · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2-Pro_img2videoqwen-vidu-viduq2-pro_img2video독립 모델vidu · 정식2025-09-24고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2-Pro_start-end2videoqwen-vidu-viduq2-pro_start-end2video독립 모델vidu · 정식2025-09-24고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2-Turbo_img2videoqwen-vidu-viduq2-turbo_img2video독립 모델vidu · 정식2025-09-24고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2-Turbo_start-end2videoqwen-vidu-viduq2-turbo_start-end2video독립 모델vidu · 정식2025-09-24고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-5 Codexgpt-5-codex독립 모델OpenAI · 정식2025-09-23최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated59.1가성비 · Estimated53.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 59.5 (estimated) · 0/19 observed / reasoning 63.0 (estimated) · 0/13 observed / writing 54.7 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.2 · 7 benchmarksReference: gpt-5.2-codex · 7 benchmarksReference: gpt-5 · 5 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당250,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

qwen-SiliconFlow DeepSeek-V3.1-Terminusqwen-siliconflow-deepseek-v3.1-terminus독립 모델DeepSeek · 정식2025-09-22고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.9가성비 · Estimated46.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Vanchin/DeepSeek-V3.1-Terminusqwen-vanchin-deepseek-v3.1-terminus독립 모델DeepSeek · 정식2025-09-22고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.9가성비 · Estimated46.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3-TTS-Flashqwen3-tts-flash독립 모델Alibaba Cloud · 정식2025-09 (공식 월 정보만 제공)사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated74.4가성비 · Estimated82.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4qwen-tts-public · Official

오디오 · 10,000자당219,179 포인트

Chirp 3-100chirp-3독립 모델Google · 정식2025-08-29고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT Realtimegpt-realtime독립 모델OpenAI · 정식2025-08-28전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated59.9가성비 · Estimated41.9

Estimate basis

GPT-5gpt-5독립 모델OpenAI · 정식2025-08-07최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated58.7가성비 · Estimated53.6

5/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 54.6 (estimated) · 2/19 observed / reasoning 65.6 (estimated) · 1/13 observed / writing 56.0 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.5-preview-2025-02-27 · 7 benchmarksReference: gpt-5.2 · 7 benchmarksReference: gpt-5.2-codex · 7 benchmarksReference: gpt-5.4 · 1 benchmarksarena · Independent arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent livebench · Independent

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당250,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

GPT-5 Minigpt-5-mini독립 모델OpenAI · 정식2025-08-07최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated43.1가성비 · Estimated49.4

4/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 39.6 (estimated) · 1/19 observed / reasoning 48.9 (estimated) · 1/13 observed / writing 40.7 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 7 benchmarksReference: gpt-5.4-mini · 7 benchmarksReference: gpt-5.1-codex-mini · 1 benchmarksReference: gpt-5.4 · 1 benchmarksarena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당500,000 포인트캐시 적중 사용량 · 백만 Token당50,000 포인트출력 사용량 · 백만 Token당4,000,000 포인트

GPT-5 Nanogpt-5-nano독립 모델OpenAI · 정식2025-08-07최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated44.4가성비 · Estimated57.0

4/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 43.4 (estimated) · 1/19 observed / reasoning 50.3 (estimated) · 1/13 observed / writing 39.5 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-nano-2025-04-14 · 7 benchmarksReference: gpt-5.4-nano · 7 benchmarksReference: gpt-5 · 1 benchmarksReference: gpt-5.4 · 1 benchmarksarena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당100,000 포인트캐시 적중 사용량 · 백만 Token당10,000 포인트출력 사용량 · 백만 Token당800,000 포인트

GPT-OSS 120B-100openai-gpt-oss-120b-maas-001독립 모델OpenAI · 정식2025-08-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated57.3가성비 · Estimated55.1

4/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 57.4 (estimated) · 1/19 observed / reasoning 63.0 (estimated) · 1/13 observed / writing 51.5 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksReference: deepseek-v4-pro-0813 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3-Coder-Flashqwen3-coder-flash독립 모델Qwen · 정식2025-08-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated58.2가성비 · Estimated64.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

Input 0 – 32,000 Token입력 사용량 · 백만 Token당273,973 포인트캐시 적중 사용량 · 백만 Token당54,795 포인트출력 사용량 · 백만 Token당1,095,890 포인트

Input 32,001 – 128,000 Token입력 사용량 · 백만 Token당410,959 포인트캐시 적중 사용량 · 백만 Token당82,192 포인트출력 사용량 · 백만 Token당1,643,836 포인트

Input 128,001 – 256,000 Token입력 사용량 · 백만 Token당684,932 포인트캐시 적중 사용량 · 백만 Token당136,986 포인트출력 사용량 · 백만 Token당2,739,726 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당1,369,863 포인트캐시 적중 사용량 · 백만 Token당273,973 포인트출력 사용량 · 백만 Token당6,849,315 포인트

Qwen-Flashqwen-flash독립 모델Qwen · 정식2025-08-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated59.4가성비 · Estimated70.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 55.4 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 8 benchmarksReference: qwen3.5-flash · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 128,000 Token입력 사용량 · 백만 Token당41,096 포인트캐시 적중 사용량 · 백만 Token당8,219 포인트출력 사용량 · 백만 Token당410,959 포인트

Input 128,001 – 256,000 Token입력 사용량 · 백만 Token당164,384 포인트캐시 적중 사용량 · 백만 Token당32,877 포인트출력 사용량 · 백만 Token당1,643,836 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당328,767 포인트캐시 적중 사용량 · 백만 Token당65,753 포인트출력 사용량 · 백만 Token당3,287,671 포인트

doubao-glm-4-5-airdoubao-glm-4-5-air-20250728고정 스냅샷 / 버전 · 유일한 진입점Zhipu AI · 정식2025-07-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GLM-4.5-Airglm-4.5-air독립 모델Zhipu · 정식2025-07-27최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated59.3가성비 · Estimated64.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 57.9 (estimated) · 0/19 observed / reasoning 64.2 (estimated) · 0/13 observed / writing 55.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: glm-4.5-air · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 31,999 Token · Output 0 – 199 Token입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당43,836 포인트출력 사용량 · 백만 Token당547,945 포인트

Input 0 – 31,999 Token · Output 200+ Token입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당43,836 포인트출력 사용량 · 백만 Token당1,643,836 포인트

Input 32,000+ Token입력 사용량 · 백만 Token당328,767 포인트캐시 적중 사용량 · 백만 Token당65,753 포인트출력 사용량 · 백만 Token당2,191,781 포인트

qwen-pixverse/pixverse-lipsyncqwen-pixverse-pixverse-lipsync독립 모델pixverse · 정식2025-07-14고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Plusqwen-plus독립 모델Qwen · 정식2025-06-24최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated44.2가성비 · Estimated55.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 45.6 (estimated) · 0/19 observed / reasoning 46.7 (estimated) · 0/13 observed / writing 40.3 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-plus · 8 benchmarksReference: qwen-plus-0125 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

Input 0 – 128,000 Token입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당43,836 포인트출력 사용량 · 백만 Token당547,945 포인트

Input 128,001 – 256,000 Token입력 사용량 · 백만 Token당657,534 포인트캐시 적중 사용량 · 백만 Token당131,507 포인트출력 사용량 · 백만 Token당5,479,452 포인트

Input 256,001+ Token입력 사용량 · 백만 Token당1,315,068 포인트캐시 적중 사용량 · 백만 Token당263,014 포인트출력 사용량 · 백만 Token당13,150,685 포인트

qwen-SiliconFlow DeepSeek-R1-0528qwen-siliconflow-deepseek-r1-0528독립 모델DeepSeek · 정식2025-05-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated43.2가성비 · Estimated45.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Mistral OCR 2505-100mistralai-mistral-ocr-2505-001독립 모델Mistral AI · 정식2025-05-22고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated81.8가성비 · Estimated72.3

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mistral-ocr-4-0 · 81.8Reference: cohere-parse-v5 · 79.2Reference: mistral-document-ai-2512 · 96.1Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.mistral-ocr · Official cohere-parsebench · Official mistral-ocr · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

mistral-medium-2505mistral-medium-2505독립 모델Mistral AI · 정식2025-05-07고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated54.1가성비 · Estimated52.9

Version comparison review4/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 47.4 (estimated) · 1/19 observed / reasoning 59.7 (estimated) · 1/13 observed / writing 55.3 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksReference: deepseek-v4-pro-0813 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.Review: below mistral-medium-3-5 by 6.5 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Phi-4 Reasoningphi-4-reasoning독립 모델Microsoft · 정식2025-04-30최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated63.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당250,000 포인트출력 사용량 · 백만 Token당1,000,000 포인트

o3o3독립 모델OpenAI · 정식2025-04-16최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.8가성비 · Estimated56.1

7/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 18%coding 57.9 (estimated) · 1/19 observed / reasoning 70.4 (estimated) · 4/13 observed / writing 60.0 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: o3-2025-04-16 · 4 benchmarksReference: smaug-flash · 3 benchmarksReference: deepseek-v4-pro-0813 · 2 benchmarksReference: DeepSeek V4 Pro (High) (0813) · 1 benchmarksopenai-simple-evals · Official microsoft-mai-report · Official mistral-codestral · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당4,000,000 포인트캐시 적중 사용량 · 백만 Token당1,000,000 포인트출력 사용량 · 백만 Token당16,000,000 포인트

Veo 2.0 Generate 001-100veo-2.0-generate-001독립 모델Google · 정식2025-04-09고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Llama 4 Maverick 17B 128Ellama-4-maverick-17b-128e-instruct-fp8독립 모델Meta · 정식2025-04-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated60.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당500,000 포인트출력 사용량 · 백만 Token당2,000,000 포인트

Llama 4 Maverick 17B 128E-100llama-4-maverick-17b-128e-instruct-maas-100독립 모델Meta · 정식2025-04-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Llama 4 Scout 17B 16E Instruct-100meta-llama-4-scout-17b-16e-instruct-maas-001독립 모델Meta · 정식2025-04-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated53.1가성비 · Estimated52.2

4/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 55.1 (estimated) · 1/19 observed / reasoning 56.7 (estimated) · 1/13 observed / writing 47.4 (estimated) · 2/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksReference: deepseek-v4-pro-0813 · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-SiliconFlow DeepSeek-V3-0324qwen-siliconflow-deepseek-v3-0324독립 모델DeepSeek · 정식2025-03-24고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated39.9가성비 · Estimated42.9

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 36.0 (estimated) · 0/19 observed / reasoning 48.3 (estimated) · 0/13 observed / writing 35.5 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3-coder-480b-a35b-instruct · 5 benchmarksReference: qwen3-30b-a3b · 4 benchmarksReference: qwen3-32b · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.Review: below qwen-deepseek-r1-0528 by 4.6 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

GPT-4o Mini Transcribegpt-4o-mini-transcribe독립 모델OpenAI · 정식2025-03-20사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.4가성비 · Estimated94.0

Estimate basis

입력 사용량 · 오디오 · 백만 Token당6,000,000 포인트입력 사용량 · 백만 Token당2,500,000 포인트출력 사용량 · 백만 Token당10,000,000 포인트

GPT-4o Mini TTSgpt-4o-mini-tts독립 모델OpenAI · 정식2025-03-20사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated59.8가성비 · Estimated56.9

Estimate basis

입력 사용량 · 백만 Token당1,200,000 포인트출력 사용량 · 오디오 · 백만 Token당24,000,000 포인트

GPT-4o Transcribegpt-4o-transcribe독립 모델OpenAI · 정식2025-03-20사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능91.4가성비64.0

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 18%qwen-asr-report · Independent

입력 사용량 · 오디오 · 백만 Token당12,000,000 포인트입력 사용량 · 백만 Token당5,000,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

mistral-small-2503mistral-small-2503독립 모델Mistral AI · 정식2025-03-17고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.0가성비—

Shared exact-model performance from mistral-small-2503; coverage refers to public model benchmarks, not tests of every platform deployment.0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.7 (estimated) · 0/19 observed / reasoning 59.2 (estimated) · 0/13 observed / writing 52.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: mistral-small-2506 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Mistral Small 3.1-100mistral-small-2503-100독립 모델Google · 미리보기2025-03-17최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.0가성비 · Estimated64.3

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.7 (estimated) · 0/19 observed / reasoning 59.2 (estimated) · 0/13 observed / writing 52.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: mistral-small-2506 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당200,000 포인트출력 사용량 · 백만 Token당600,000 포인트

Text Embedding Large Exp 03 07-100text-embedding-large-exp-03-07독립 모델Google · 정식2025-03-07고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-DeepSeek-R1-Distill-Qwen-1.5Bqwen-deepseek-r1-distill-qwen-1.5b독립 모델DeepSeek · 정식2025-01-20고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated43.2가성비 · Estimated45.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Codestral 2501codestral-2501독립 모델Mistral AI · 정식2025-01-13전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated60.1

9/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 18%coding 56.8 (estimated) · 9/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksReference: smaug-flash · 3 benchmarksmistral-codestral · Official microsoft-mai-report · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당600,000 포인트출력 사용량 · 백만 Token당1,800,000 포인트

Ministral-3Bministral-3b독립 모델Mistral AI · 정식2024-10-16고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated49.2가성비 · Estimated49.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 53.8 (estimated) · 0/19 observed / reasoning 50.7 (estimated) · 0/13 observed / writing 43.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: ministral-8b-2410 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: smaug-flash · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Longqwen-long독립 모델Qwen · 정식2024-05-20최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated52.5가성비 · Estimated62.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 48.7 (estimated) · 0/19 observed / reasoning 58.7 (estimated) · 0/13 observed / writing 50.2 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-27b · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen-glm-4.6 · 5 benchmarksReference: qwen-glm-4.5 · 4 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당136,986 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당547,945 포인트

Text Embedding 3 Largetext-embedding-3-large독립 모델OpenAI · 정식2024-01-25최근 가격 확인
확인 대기 (UTC+8)

성능64.6가성비45.2

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 15%openai-embedding · Official

임베딩 입력 · 백만 Token당260,000 포인트

Text Embedding 3 Smalltext-embedding-3-small독립 모델OpenAI · 정식2024-01-25최근 가격 확인
확인 대기 (UTC+8)

성능62.3가성비71.3

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 15%openai-embedding · Official

임베딩 입력 · 백만 Token당40,000 포인트

Cohere-embed-v3-englishcohere-embed-v3-english독립 모델Cohere · 정식2023-11-02고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Cohere-embed-v3-multilingualcohere-embed-v3-multilingual독립 모델Cohere · 정식2023-11-02고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Text Embedding Ada 002text-embedding-ada-002독립 모델OpenAI · 정식2022-12-15최근 가격 확인
확인 대기 (UTC+8)

성능61.0가성비46.6

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 15%openai-embedding · Official

임베딩 입력 · 백만 Token당200,000 포인트

Qwen-Image-2.1-Turboqwen-image-2.1-turbo독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-14고정 단가 없이 실제 서비스에 따라 과금

성능—가성비—

미평가

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Chirp 2-100chirp-2-001독립 모델Google · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

语音生成CosyVoice-v3.5-flash大模型cosyvoice-v3.5-flash독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated75.4가성비 · Estimated67.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Gemini-pro · 82.6Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-tts-public · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

语音生成CosyVoice-v3.5-plus大模型cosyvoice-v3.5-plus독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated75.4가성비 · Estimated67.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Gemini-pro · 82.6Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-tts-public · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

doubao-GLM-5.3-Flashdoubao-glm-5-3-flash-260828고정 스냅샷 / 버전 · 유일한 진입점Zhipu · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated63.3

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당63,014 포인트출력 사용량 · 백만 Token당767,123 포인트

qwen2-5-72bdoubao-qwen2-5-72b-20240919고정 스냅샷 / 버전 · 유일한 진입점Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen3-0-6bdoubao-qwen3-0-6b-20250429고정 스냅샷 / 버전 · 유일한 진입점Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen3-14bdoubao-qwen3-14b-20250429고정 스냅샷 / 버전 · 유일한 진입점Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen3-32bdoubao-qwen3-32b-20250429고정 스냅샷 / 버전 · 유일한 진입점Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen3-8bdoubao-qwen3-8b-20250429고정 스냅샷 / 버전 · 유일한 진입점Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

doubao-seed-translationdoubao-seed-translation-250915고정 스냅샷 / 버전 · 유일한 진입점Doubao · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

doubao-smart-routerdoubao-smart-router-250928고정 스냅샷 / 버전 · 유일한 진입점Doubao · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated50.0가성비 · Estimated50.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: o1-mini · 50.0Reference: qwen3.6-plus · 50.2Reference: gemini-2.0-flash-001 · 50.3Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

FaceChain人物图像检测facechain-facedetect독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.5가성비 · Estimated46.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.7-image-pro · 44.1Reference: krea-2-medium · 45.0Reference: reve-v1.1 · 43.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Fun-ASR语音识别fun-asr독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Fun-ASR实时语音识别fun-asr-realtime독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

音乐生成 Previewfun-music-preview독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

音乐生成fun-music-v1독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Grok 4.20 Non-Reasoning-100grok-4-20-non-reasoning-100독립 모델xAI · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비—

Estimate basis

입력 사용량 · 백만 Token당2,500,000 포인트캐시 적중 사용량 · 백만 Token당400,000 포인트출력 사용량 · 백만 Token당5,000,000 포인트

Grok 4.20 Reasoning-100grok-4-20-reasoning-100독립 모델xAI · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비—

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Grok 4.3-100grok-4.3-100독립 모델xAI · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.8가성비—

Shared exact-model performance from grok-4.3; coverage refers to public model benchmarks, not tests of every platform deployment.12/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 38.8 (estimated) · 4/19 observed / reasoning 48.3 (estimated) · 4/13 observed / writing 47.2 (estimated) · 4/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: gpt-4.1 · 2 benchmarksReference: o1-mini · 2 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent livebench · Independent microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-HappyHorse-1.0-Video-Edithappyhorse-1.0-video-edit독립 모델HappyHorse · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated80.9가성비 · Estimated71.6

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: happyhorse-1.0 · 80.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

图像擦除补全image-erase-completion독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.5가성비 · Estimated46.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.7-image-pro · 44.1Reference: krea-2-medium · 45.0Reference: reve-v1.1 · 43.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

人物实例分割image-instance-segmentation독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.5가성비 · Estimated46.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.7-image-pro · 44.1Reference: krea-2-medium · 45.0Reference: reve-v1.1 · 43.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Codestral 2-100mistralai-codestral-2-001독립 모델Mistral AI · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Mistral Medium 3-100mistralai-mistral-medium-3-001독립 모델Mistral AI · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Paraformer语音识别-8k-v2paraformer-8k-v2독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Paraformer实时语音识别-8k-v2paraformer-realtime-8k-v2독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Paraformer实时语音识别-v2paraformer-realtime-v2독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Paraformer语音识别-v2paraformer-v2독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3-ASR-Flash-Filetransqwen3-asr-flash-filetrans독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen3-ASR-Flash-Realtimeqwen3-asr-flash-realtime독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.6가성비 · Estimated51.8

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.1-ASR-Flashqwen-audio-3.1-asr-flash독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.1-ASR-Flash-Filetransqwen-audio-3.1-asr-flash-filetrans독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.1-ASR-Flash-Messageqwen-audio-3.1-asr-flash-message독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.1-ASR-Flash-Streamingqwen-audio-3.1-asr-flash-streaming독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated91.4가성비 · Estimated79.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-transcribe · 91.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-asr-report · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.1-TTS-Flashqwen-audio-3.1-tts-flash독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated74.4가성비 · Estimated67.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-tts-public · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Audio-3.1-TTS-Nextqwen-audio-3.1-tts-next독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated74.4가성비 · Estimated67.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.qwen-tts-public · Official

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Doc-Turboqwen-doc-turbo독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated48.7가성비 · Estimated49.1

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 46.8 (estimated) · 0/19 observed / reasoning 52.3 (estimated) · 0/13 observed / writing 47.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-27b · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3.6-plus · 7 benchmarksReference: qwen-deepseek-r1 · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당164,384 포인트캐시 적중 사용량 · 백만 Token당32,877 포인트출력 사용량 · 백만 Token당273,973 포인트

Qwen-Flash-Characterqwen-flash-character독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated59.9가성비 · Estimated70.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 57.1 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 8 benchmarksReference: qwen3.5-flash · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당68,493 포인트캐시 적중 사용량 · 백만 Token당13,699 포인트출력 사용량 · 백만 Token당410,959 포인트

qwen-Kling Video 3.0 Turboqwen-kling-kling-v3-turbo-video-generation독립 모델kling · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-speech-02-hdqwen-minimax-speech-02-hd독립 모델mini-max · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated32.9가성비 · Estimated38.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 32.7 (estimated) · 0/19 observed / reasoning 39.5 (estimated) · 0/13 observed / writing 26.5 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen2-72b-instruct · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-speech-02-turboqwen-minimax-speech-02-turbo독립 모델mini-max · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated34.1가성비 · Estimated38.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 34.1 (estimated) · 0/19 observed / reasoning 39.8 (estimated) · 0/13 observed / writing 28.4 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3.6-plus · 7 benchmarksReference: qwen2-72b-instruct · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-speech-2.8-hdqwen-minimax-speech-2.8-hd독립 모델mini-max · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.1가성비 · Estimated45.9

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 38.0 (estimated) · 0/19 observed / reasoning 53.2 (estimated) · 0/13 observed / writing 41.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3-235b-a22b-instruct-2507 · 4 benchmarksReference: qwen3-235b-a22b-thinking-2507 · 4 benchmarksReference: qwen3-30b-a3b-instruct-2507 · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-speech-2.8-turboqwen-minimax-speech-2.8-turbo독립 모델mini-max · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated46.9가성비 · Estimated47.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 40.0 (estimated) · 0/19 observed / reasoning 56.2 (estimated) · 0/13 observed / writing 44.6 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3.6-plus · 7 benchmarksReference: qwen3-30b-a3b-instruct-2507 · 4 benchmarksReference: qwen3-max-2025-09-23 · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-MT-Flashqwen-mt-flash독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated59.4가성비 · Estimated56.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 55.6 (estimated) · 0/19 observed / reasoning 62.8 (estimated) · 0/13 observed / writing 59.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 8 benchmarksReference: qwen3.5-flash · 5 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-MT-Imageqwen-mt-image독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated34.1가성비 · Estimated38.9

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen-image · 20.0Reference: qwen-image-prompt-extend · 23.8Reference: qwen-image-2.1 · 73.8Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-MT-Liteqwen-mt-lite독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.3가성비 · Estimated54.4

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 50.2 (estimated) · 0/19 observed / reasoning 60.6 (estimated) · 0/13 observed / writing 58.1 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.7-max · 8 benchmarksReference: qwen3.8-max · 8 benchmarksReference: qwen3.6-plus · 7 benchmarksReference: qwen-glm-4.6 · 5 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-MT-Plusqwen-mt-plus독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated46.7가성비 · Estimated47.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 46.9 (estimated) · 0/19 observed / reasoning 50.1 (estimated) · 0/13 observed / writing 43.2 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: qwen3.6-plus · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen-plus-0125 · 4 benchmarksReference: qwen2.5-plus-1127 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-PixVerse-V6-r2vqwen-pixverse-pixverse-v6-r2v독립 모델pixverse · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Plus-Characterqwen-plus-character독립 모델qwen-domain-model · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated46.9가성비 · Estimated57.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 47.0 (estimated) · 0/19 observed / reasoning 50.2 (estimated) · 0/13 observed / writing 43.4 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: qwen3.6-plus · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen-plus-0125 · 4 benchmarksReference: qwen2.5-plus-1127 · 4 benchmarksReference: gpt-4.1-mini · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당219,178 포인트캐시 적중 사용량 · 백만 Token당43,836 포인트출력 사용량 · 백만 Token당547,945 포인트

qwen-stepfun/step-5-previewqwen-stepfun-step-5-preview독립 모델StepFun · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated30.7

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당1,917,808 포인트캐시 적중 사용량 · 백만 Token당95,890 포인트출력 사용량 · 백만 Token당5,479,452 포인트

qwen-unisound/unisound-u2qwen-unisound-unisound-u2독립 모델unisound · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated32.9가성비 · Estimated38.0

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Vanchin/DeepSeek-R1qwen-vanchin-deepseek-r1독립 모델DeepSeek · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated43.2가성비 · Estimated45.2

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Vanchin/DeepSeek-V3qwen-vanchin-deepseek-v3독립 모델DeepSeek · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated42.6가성비 · Estimated44.8

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 37.2 (estimated) · 0/19 observed / reasoning 51.7 (estimated) · 0/13 observed / writing 39.0 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen3-235b-a22b-no-thinking · 4 benchmarksReference: qwen3-next-80b-a3b-thinking · 4 benchmarksReference: qwen3-vl-235b-a22b-thinking · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.Review: below qwen-deepseek-r1-0528 by 1.9 points. Check matching benchmarks, configurations and estimates; this advisory does not alter the score.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-vanchin/deepseek-v4-proqwen-vanchin-deepseek-v4-pro독립 모델DeepSeek · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated71.1가성비 · Estimated64.8

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 67.0 (estimated) · 0/19 observed / reasoning 75.3 (estimated) · 0/13 observed / writing 70.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen3.8-max · 13 benchmarksReference: mai-thinking-1 · 12 benchmarksReference: qwen3.8-27b · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-flash-next · 8 benchmarksReference: gpt-4.1-mini · 3 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Vidu-image-lite_reference2imageqwen-vidu-vidu-image-lite_reference2image독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated66.4가성비 · Estimated61.5

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Vidu-image-pro_reference2imageqwen-vidu-vidu-image-pro_reference2image독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated66.4가성비 · Estimated61.5

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Vidu-image_reference2imageqwen-vidu-vidu-image_reference2image독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated73.8가성비 · Estimated66.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen-image-2.1 · 73.8Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2-fast_reference2imageqwen-vidu-viduq2-fast_reference2image독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated66.4가성비 · Estimated61.5

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ2-Pro_reference2imageqwen-vidu-viduq2-pro_reference2image독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated66.4가성비 · Estimated61.5

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Ad_reference2videoqwen-vidu-viduq3-ad_reference2video독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Drama_reference2videoqwen-vidu-viduq3-drama_reference2video독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-fast_reference2imageqwen-vidu-viduq3-fast_reference2image독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated85.0가성비 · Estimated74.5

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: qwen-image-3.0-pro · 85.0Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ViduQ3-Pro-fast_img2videoqwen-vidu-viduq3-pro-fast_img2video독립 모델vidu · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Qwen-Voice-Enrollmentqwen-voice-enrollment독립 서비스 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 45.7 (estimated) · 0/19 observed / reasoning 56.4 (estimated) · 0/13 observed / writing 50.9 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: qwen3.8-27b · 8 benchmarksReference: qwen3.6-27b · 7 benchmarksReference: qwen-deepseek-r1 · 4 benchmarksReference: qwen-deepseek-r1-0528 · 4 benchmarksComparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.microsoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-ZHIPU/GLM-5.3-FlashXqwen-zhipu-glm-5.3-flashx독립 모델Zhipu AI · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated41.0가성비 · Estimated37.8

Version comparison review0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당547,945 포인트캐시 적중 사용량 · 백만 Token당156,164 포인트출력 사용량 · 백만 Token당1,917,808 포인트

Sambert语音合成-Bethsambert-beth-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Bettysambert-betty-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Briansambert-brian-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Callysambert-cally-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Camilasambert-camila-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Cindysambert-cindy-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Clarasambert-clara-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Donnasambert-donna-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Evasambert-eva-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Hannasambert-hanna-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Indahsambert-indah-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Perlasambert-perla-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-Waansambert-waan-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知厨sambert-zhichu-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知达sambert-zhida-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知德sambert-zhide-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知飞sambert-zhifei-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知柜sambert-zhigui-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.5가성비 · Estimated46.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.7-image-pro · 44.1Reference: krea-2-medium · 45.0Reference: reve-v1.1 · 43.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知浩sambert-zhihao-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知佳sambert-zhijia-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知婧sambert-zhijing-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知伦sambert-zhilun-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知猫sambert-zhimao-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知妙（多情感）sambert-zhimiao-emo-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知茗sambert-zhiming-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知墨sambert-zhimo-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知楠sambert-zhinan-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知娜sambert-zhina-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知倩sambert-zhiqian-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知琪sambert-zhiqi-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知茹sambert-zhiru-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知硕sambert-zhishuo-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知树sambert-zhishu-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知莎sambert-zhistella-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知婷sambert-zhiting-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知薇sambert-zhiwei-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知祥sambert-zhixiang-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知笑sambert-zhixiao-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知雅sambert-zhiya-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知晔sambert-zhiye-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知颖sambert-zhiying-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知媛sambert-zhiyuan-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

Sambert语音合成-知悦sambert-zhiyue-v1독립 음색 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

鞋靴模特shoemodel-v1독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.5가성비 · Estimated46.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.7-image-pro · 44.1Reference: krea-2-medium · 45.0Reference: reve-v1.1 · 43.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

语音识别热词speech-biasing독립 서비스 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

通用文本向量-async-v1text-embedding-async-v1독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

通用文本向量-async-v2text-embedding-async-v2독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

意图分类模型tongyi-intent-detect-v3독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

伶鹊-对话分析-flashtongyi-xiaomi-analysis-flash독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당54,795 포인트캐시 적중 사용량 · 백만 Token당10,959 포인트출력 사용량 · 백만 Token당109,589 포인트

伶鹊-对话分析-protongyi-xiaomi-analysis-pro독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

입력 사용량 · 백만 Token당273,973 포인트캐시 적중 사용량 · 백만 Token당54,795 포인트출력 사용량 · 백만 Token당739,726 포인트

Translate LLM-100translate-llm독립 모델Google · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

声动人像VideoRetalkvideoretalk독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

视频风格重绘video-style-transform독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.0가성비 · Estimated50.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: seedance-v1.5-pro · 53.2Reference: wan2.6-i2v · 53.2Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

虚拟模特V2virtualmodel-v2독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated44.5가성비 · Estimated46.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.7-image-pro · 44.1Reference: krea-2-medium · 45.0Reference: reve-v1.1 · 43.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

大模型声音复刻及声音设计voice-enrollment독립 서비스 항목Qwen · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated56.6가성비 · Estimated54.6

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.2-Animate-Mixwan2.2-animate-mix독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.2-Animate-Movewan2.2-animate-move독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.2-KF2V-Flashwan2.2-kf2v-flash독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-通义万相2.2-数字人-S2Vwan2.2-s2v독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.2-S2V-Detectwan2.2-s2v-detect독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.2-T2I-Flashwan2.2-t2i-flash독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.2-T2I-Pluswan2.2-t2i-plus독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.5-I2I-Previewwan2.5-i2i-preview독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated52.7가성비 · Estimated51.9

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-i2v-preview · 55.3Reference: wan2.5-t2v-preview · 48.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.5-T2I-Previewwan2.5-t2i-preview독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated51.5가성비 · Estimated51.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.5-t2v-preview · 48.9Reference: wan2.5-i2v-preview · 55.3Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.6-R2V-Flashwan2.6-r2v-flash독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated55.3가성비 · Estimated53.7

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan2.6-i2v · 53.2Reference: wan2.6-t2v · 57.4Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.0-T2I-Turbowanx2.0-t2i-turbo독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.1-ImageEditwanx2.1-imageedit독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.1-KF2V-Pluswanx2.1-kf2v-plus독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.1-T2I-Pluswanx2.1-t2i-plus독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.1-T2I-Turbowanx2.1-t2i-turbo독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-Wan2.1-VACE-Pluswanx2.1-vace-plus독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-创意海报生成wanx-poster-generation-v1독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-万相-涂鸦作画wanx-sketch-to-image-lite독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated34.8가성비 · Estimated39.4

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Reference: wan2.5-i2v-preview · 55.3Reference: wan2.5-t2v-preview · 48.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-人像风格重绘wanx-style-repaint-v1독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-wanx-t2iwanx-v1독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-虚拟模特wanx-virtualmodel독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated36.9가성비 · Estimated40.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Reference: wan2.5-i2v-preview · 55.3Reference: wan2.5-t2v-preview · 48.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

qwen-万相-图像局部重绘wanx-x-painting독립 모델Wan · 정식목록 추가 (출시일 확인 대기): 2026-10-05고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated37.0가성비 · Estimated40.9

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: wan-v2.2-a14b · 11.7Reference: wan2.5-i2v-preview · 55.3Reference: wan2.5-t2v-preview · 48.9Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.arena · Independent arena · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

语音生成cosyvoice-v2大模型cosyvoice-v2독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated75.4가성비 · Estimated52.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Gemini-pro · 82.6qwen-tts-public · Official

오디오 · 10,000자당547,946 포인트

语音生成CosyVoice-v3-flash大模型cosyvoice-v3-flash독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated75.4가성비 · Estimated75.5

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Gemini-pro · 82.6qwen-tts-public · Official

오디오 · 10,000자당273,973 포인트

语音生成CosyVoice-v3-plus大模型cosyvoice-v3-plus독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated75.4가성비 · Estimated52.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3TTS-25Hz-1.7B-CustomVoice · 75.4Reference: Qwen3TTS-12Hz-1.7B-CustomVoice · 73.4Reference: Gemini-pro · 82.6qwen-tts-public · Official

오디오 · 10,000자당547,946 포인트

GUI-Plusgui-plus독립 모델Qwen · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated56.6가성비 · Estimated61.7

0/41 fixed-panel observations; remaining cells are marked estimates

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%coding 56.8 (estimated) · 0/19 observed / reasoning 60.2 (estimated) · 0/13 observed / writing 52.8 (estimated) · 0/9 observedEvery chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mai-thinking-1 · 12 benchmarksReference: codestral-2501 · 9 benchmarksReference: gpt-4.1-mini · 3 benchmarksReference: gpt-4o-2024-05-13 · 3 benchmarksReference: o1-mini · 3 benchmarksReference: o1-preview · 3 benchmarksmicrosoft-mai-report · Official mistral-codestral · Official openai-simple-evals · Official arena · Independent arena · Independent arena · Independent livebench · Independent

입력 사용량 · 백만 Token당410,959 포인트캐시 적중 사용량 · 백만 Token당0 포인트출력 사용량 · 백만 Token당1,232,877 포인트

通用多模态向量multimodal-embedding-v1독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.3가성비 · Estimated48.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: text-embedding-3-small · 62.3Reference: text-embedding-v3 · 63.4Reference: text-embedding-ada-002 · 61.0openai-embedding · Official aliyun-embedding · Official

임베딩 입력 · 백만 Token당191,781 포인트입력 사용량 · 이미지 · 백만 Token당246,576 포인트비디오 입력 · 백만 Token당246,576 포인트

Qwen2.5-VL-Embeddingqwen2.5-vl-embedding독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.3가성비 · Estimated48.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: text-embedding-3-small · 62.3Reference: text-embedding-v3 · 63.4Reference: text-embedding-ada-002 · 61.0openai-embedding · Official aliyun-embedding · Official

임베딩 입력 · 백만 Token당191,781 포인트입력 사용량 · 이미지 · 백만 Token당493,151 포인트

Qwen3-ASR-Flashqwen3-asr-flash독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.4가성비 · Estimated94.0

Estimate basis

음성 전사 · 초당61 포인트

千问3-Rerankqwen3-rerank독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated72.5가성비 · Estimated80.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3-Reranker-4B · 73.9Reference: Qwen3-Reranker-8B · 74.2Reference: Qwen3-Reranker-0.6B · 68.8qwen-rerank-public · Official

입력 사용량 · 백만 Token당136,987 포인트

Qwen3-VL-Embeddingqwen3-vl-embedding독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.3가성비 · Estimated48.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: text-embedding-3-small · 62.3Reference: text-embedding-v3 · 63.4Reference: text-embedding-ada-002 · 61.0openai-embedding · Official aliyun-embedding · Official

임베딩 입력 · 백만 Token당191,781 포인트입력 사용량 · 이미지 · 백만 Token당493,151 포인트

Qwen3-VL-Rerankqwen3-vl-rerank독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated72.5가성비 · Estimated80.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 24%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: Qwen3-Reranker-4B · 73.9Reference: Qwen3-Reranker-8B · 74.2Reference: Qwen3-Reranker-0.6B · 68.8qwen-rerank-public · Official

입력 사용량 · 백만 Token당136,987 포인트비디오 입력 · 백만 Token당493,154 포인트

사용자가 확인한 비디오 입력 임시 책임 가격입니다. 공급자는 공식 비디오 가격과 프레임-Token 환산 기준을 공개하지 않았습니다. 공개 카탈로그 표시와 검수 예약에만 사용하며 공급자 실제 비용으로 기록하거나 기존 청구를 변경하지 않습니다.

Qwen-Image-2.0-Proqwen-image-2.0-pro독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated66.4가성비 · Estimated46.5

Estimate basis

출력 사용량 · 이미지 · 이미지당136,987 포인트

Qwen-Image-3.0-Proqwen-image-3.0-pro독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능85.0가성비59.5

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

이미지 입력 사용량 1K · 이미지당5,480 포인트이미지 입력 사용량 2K · 이미지당5,480 포인트이미지 출력 사용량 1K · 이미지당68,494 포인트이미지 출력 사용량 2K · 이미지당136,987 포인트

QwenVL-OCRqwen-vl-ocr독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated81.8가성비 · Estimated72.3

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: mistral-ocr-4-0 · 81.8Reference: cohere-parse-v5 · 79.2Reference: mistral-document-ai-2512 · 96.1Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.mistral-ocr · Official cohere-parsebench · Official mistral-ocr · Official

입력 사용량 · 백만 Token당82,192 포인트출력 사용량 · 백만 Token당136,987 포인트

通用文本向量-v1text-embedding-v1독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능58.3가성비45.3

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 18%aliyun-embedding · Official

임베딩 입력 · 백만 Token당191,781 포인트

通用文本向量-v2text-embedding-v2독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능60.1가성비46.6

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 18%aliyun-embedding · Official

임베딩 입력 · 백만 Token당191,781 포인트

通用文本向量-v3text-embedding-v3독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능63.4가성비53.9

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 18%aliyun-embedding · Official

임베딩 입력 · 백만 Token당136,987 포인트

通用文本向量-v4text-embedding-v4독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능71.6가성비59.6

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 18%aliyun-embedding · Official

임베딩 입력 · 백만 Token당136,987 포인트

视觉向量-flashtongyi-embedding-vision-flash독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.3가성비 · Estimated70.9

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: text-embedding-3-small · 62.3Reference: text-embedding-v3 · 63.4Reference: text-embedding-ada-002 · 61.0openai-embedding · Official aliyun-embedding · Official

임베딩 입력 · 백만 Token당41,096 포인트

视觉向量-plustongyi-embedding-vision-plus독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated62.3가성비 · Estimated53.1

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: text-embedding-3-small · 62.3Reference: text-embedding-v3 · 63.4Reference: text-embedding-ada-002 · 61.0openai-embedding · Official aliyun-embedding · Official

임베딩 입력 · 백만 Token당136,987 포인트

Wan2.2-I2V-Flashwan2.2-i2v-flash독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated11.7가성비 · Estimated38.2

Estimate basis

비디오 생성 480P · 초당27,398 포인트비디오 생성 720P · 초당54,795 포인트비디오 생성 1080P · 초당131,507 포인트

Wan2.2-I2V-Pluswan2.2-i2v-plus독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

비디오 생성 480P · 초당38,357 포인트비디오 생성 1080P · 초당191,781 포인트

Wan2.2-T2V-Pluswan2.2-t2v-plus독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated11.7가성비 · Estimated23.2

Estimate basis

비디오 생성 480P · 초당38,357 포인트비디오 생성 1080P · 초당191,781 포인트

Wan2.5-I2V-Previewwan2.5-i2v-preview독립 모델Alibaba Cloud · 미리보기목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능55.3가성비46.8

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

비디오 생성 480P · 초당82,192 포인트비디오 생성 720P · 초당164,384 포인트비디오 생성 1080P · 초당273,973 포인트

Wan2.5-T2V-Previewwan2.5-t2v-preview독립 모델Alibaba Cloud · 미리보기목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능48.9가성비42.3

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 45%arena · Independent

비디오 생성 480P · 초당82,192 포인트비디오 생성 720P · 초당164,384 포인트비디오 생성 1080P · 초당273,973 포인트

wan2.6-I2V-flashwan2.6-i2v-flash독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated54.9가성비 · Estimated60.3

Estimate basis

비디오 생성 720P · 초당82,192 포인트비디오 생성 1080P · 초당136,987 포인트

Wan2.1-I2V-Pluswanx2.1-i2v-plus독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated11.7가성비 · Estimated13.2

Estimate basis

비디오 생성 720P · 초당191,781 포인트

Wan2.1-I2V-Turbowanx2.1-i2v-turbo독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated11.7가성비 · Estimated34.6

Estimate basis

비디오 생성 480P · 초당65,754 포인트비디오 생성 720P · 초당65,754 포인트

Wan2.1-T2V-Pluswanx2.1-t2v-plus독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated11.7가성비 · Estimated13.2

Estimate basis

비디오 생성 720P · 초당191,781 포인트

Wan2.1-T2V-Turbowanx2.1-t2v-turbo독립 모델Alibaba Cloud · 정식목록 추가 (출시일 확인 대기): 2026-09-10최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated11.7가성비 · Estimated34.6

Estimate basis

비디오 생성 480P · 초당65,754 포인트비디오 생성 720P · 초당65,754 포인트

doubao-hitem3d-2-0-251223doubao-hitem3d-2-0-251223고정 스냅샷 / 버전 · 유일한 진입점Hitem3D · 정식목록 추가 (출시일 확인 대기): 2026-09-09최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated82.2가성비 · Estimated57.5

Estimate basis

출력 사용량 · 3D · 백만 Token당21,917,809 포인트

doubao-hyper3d-gen2-260112doubao-hyper3d-gen2-260112고정 스냅샷 / 버전 · 유일한 진입점Hyper3D · 정식목록 추가 (출시일 확인 대기): 2026-09-09최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated82.2가성비 · Estimated87.5

Estimate basis

출력 사용량 · 3D · 백만 Token당16,438,357 포인트

doubao-seedance-2-0-260128doubao-seedance-2-0-260128고정 스냅샷 / 버전 · 유일한 진입점ByteDance · 정식목록 추가 (출시일 확인 대기): 2026-09-09최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.5가성비 · Estimated68.5

Estimate basis

출력 사용량 · 비디오 생성 · 백만 Token당12,602,740 포인트

doubao-seedance-2-0-fast-260128doubao-seedance-2-0-fast-260128고정 스냅샷 / 버전 · 유일한 진입점ByteDance · 정식목록 추가 (출시일 확인 대기): 2026-09-09최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.5가성비 · Estimated70.8

Estimate basis

출력 사용량 · 비디오 생성 · 백만 Token당7,602,740 포인트

doubao-seedance-2-0-mini-260615doubao-seedance-2-0-mini-260615고정 스냅샷 / 버전 · 유일한 진입점ByteDance · 정식목록 추가 (출시일 확인 대기): 2026-09-09최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.5가성비 · Estimated75.9

Estimate basis

출력 사용량 · 비디오 생성 · 백만 Token당2,520,548 포인트

doubao-seedream-5-0-260128doubao-seedream-5-0-260128고정 스냅샷 / 버전 · 유일한 진입점ByteDance · 정식목록 추가 (출시일 확인 대기): 2026-09-09사용 가능 · 생성과 편집최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated54.4가성비 · Estimated53.4

Estimate basis

출력 사용량 · 이미지 · 이미지당60,274 포인트

GPT-4o Transcribe Diarizegpt-4o-transcribe-diarize독립 모델OpenAI · 정식목록 추가 (출시일 확인 대기): 2026-08-28사용 가능 · 음성 도구최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated91.4가성비 · Estimated64.0

Estimate basis

입력 사용량 · 오디오 · 백만 Token당12,000,000 포인트입력 사용량 · 백만 Token당5,000,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트

GPT Audiogpt-audio독립 모델OpenAI · 정식목록 추가 (출시일 확인 대기): 2026-08-28전체 기능최근 가격 확인
확인 대기 (UTC+8)

성능 · Estimated52.6가성비 · Estimated36.8

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: gpt-4o-realtime-preview-2024-12-17 · 45.3Reference: gpt-realtime-2025-08-28 · 59.9openai-realtime-public · Official

입력 사용량 · 백만 Token당5,000,000 포인트출력 사용량 · 백만 Token당20,000,000 포인트입력 사용량 · 오디오 · 백만 Token당80,000,000 포인트출력 사용량 · 오디오 · 백만 Token당160,000,000 포인트

Mistral Document AI 2512mistral-document-ai-2512독립 모델Mistral AI · 정식목록 추가 (출시일 확인 대기): 2026-08-28최근 가격 확인
확인 대기 (UTC+8)

성능96.1가성비76.1

잠정 평가

2026-10-08 16:22 (UTC+8) · Confidence 18%mistral-ocr · Official

문서 처리 · 1,000페이지당6,000,000 포인트

Model Routermodel-router독립 라우팅 서비스OpenAI · 정식목록 추가 (출시일 확인 대기): 2026-08-28고정 단가 없이 실제 서비스에 따라 과금

성능 · Estimated50.0가성비 · Estimated50.0

Estimate basis

2026-10-08 16:22 (UTC+8) · Confidence 12%Every chat model uses the same benchmark panel and weights. Missing cells are estimated from the same base version first, then nearby same-purpose models.Reference: o1-mini · 50.0Reference: qwen3.6-plus · 50.2Reference: gemini-2.0-flash-001 · 50.3Comparable price unavailable; value is estimated with a neutral cost component and is excluded from value defaults.openai-simple-evals · Official arena · Independent arena · Independent livebench · Independent

동적 요금: 라우터 입력 가산점과 실제 선택 모델의 포인트 합계.

검증된 기능 설명 · 44

Claude Fable 5.1: complete within the verified, safely available deployment scope

Five adaptive efforts, high by default. Native/shared tools, structured output, real Codex continuation, image/PDF, cache and Key-owned files were verified.

Five adaptive efforts, high by default. Native/shared tools, structured output, real Codex continuation, image/PDF, cache and Key-owned files were verified.
Code execution uses the HCD free-combination billing bridge, adding fetch restricted to a non-resolving domain while preserving client tools. Execution/extremes are not live-tested (high-cost exemption); standalone native timing metering is not claimed.
Performance71.6/value55.1 are estimates:7 direct and34 estimated metrics in the fixed41-panel, using exact LiveBench max data.

Claude Haiku 5.5：Complete within the verified deployment scope

The exact available scope is verified; optional parameters and resources retain explicit bounds.

none disables thinking; low/medium/high/xhigh/max remain their real values; medium is the default. Native between_tools is rejected for this exact model.
Native Messages and the shared Chat/Responses path, strict tool schemas, inline image/PDF, SSE and signed history passed. The actual Codex complete tools, MCP echo and saved-thread continuation passed.
Bounded hosted search/fetch with citations and one-use owned remote MCP passed and settled once. Arbitrary remote URLs and unconstrained hosted execution are not enabled.
Ordinary input, output, cache reads and 5m/1h writes have separate rates. Input including cache reads/writes above100K uses the whole-request long-context tier at5x; the threshold boundary was checked offline.
The current deployment has no Files/Batch/container/skills API. Client function tools run on the client; no new native computer/browser toolset, durable background Responses resource or native compaction pass is claimed.
1M context and HCD128000 output limits retain safe reservation/settlement. Expensive extremes were not live-tested (high-cost exemption).
Performance55.1 and value61.9 are estimated:8 directly attributed fixed-panel cells and33 estimates out of41. Benchmark effort high is separate from the medium inference default.

Claude Opus 5.5: complete verified deployment scope

Text, image/PDF, tools, thinking, cache and web access through Messages, Chat, Responses and Codex compatibility.

Adaptive thinking is always on; medium is the default. Only temperature 1 is supported; top_p/top_k are unavailable.
Verified client bridging supplies tool discovery. Native server tool search returned a safety refusal and is not claimed passed.
1M context and 128K output. Expensive limits are not live-tested (high-cost exemption).
Files, Batch, code execution and native background jobs are not exposed. URL media generation is verified; token counting accepts supported inline inputs.
Client-side compaction and same-thread continuation are verified; unpriced server compaction remains closed.
Performance and value scores await exact-model evidence.

Claude Sonnet 5.5: complete within the verified, safely available deployment scope

Completion covers the exact model’s available scope. Optional hosted capabilities, resource quantities and budgets retain explicit bounds.

none, low, medium, high, xhigh and max; high is default. none uses between_tools: no up-front thinking, while progress between tools is retained. none is never aliased to low.
Native Messages and shared Chat/Responses conversion, strict structured output, advisory task-budget beta, functions, original SSE/signed thinking history, inline image/PDF, both cache TTLs and cache reads were verified.
Actual Codex complete default tools, discovery, top-level composed-schema bridging, an in-memory MCP round trip and a persisted second turn passed. Advanced beta controls require the corresponding official header.
Basic hosted search/fetch, citations and actual-usage accounting passed. No provider-native server Tool Search pass is claimed; client discovery bridging passed.
A funded owned remote-MCP HTTPS echo and Key/project lifecycle were verified; its lease was closed. Arbitrary remote URLs, variable results and external credential management remain closed.
The offering has no Batch, Files, code execution/containers/skills or new computer/browser toolsets. Messages has no native durable background-response resource. Native compaction remains closed; HCD’s existing client context path is reused without claiming an additional Sonnet compaction-generation live test.
Official 1M context and 128K output use uniform full-context rates. Expensive extremes are not live-tested (high-cost exemption); offline limits, safe holds and settlement remain enforced.
Performance 70.6 and value 61.1 are explicitly estimated: exact-model LiveBench observations cover 7 of the fixed 41-panel metrics at max effort. Value uses the same current official points tariff.

Codestral 2501: complete within the verified dedicated deployment scope

Verified Codestral-2501 native text chat, fill-in-the-middle, prefix/suffix completion, streaming and sampling through a dedicated code-tool entry.

It is a text/code tool, not a function-tool-capable Codex main or subagent model.
Exact FIM rejects metadata, prompt_cache_key and explicit null sampling/suffix values; HCD rejects them before dispatch rather than silently dropping them.
Native responses and actual token usage are retained. Estimated scores are not derived from this interface acceptance.

Cohere Parse v5: complete within the verified dedicated deployment scope

Verified Cohere Parse v5 dedicated image parsing with text/Markdown, blocks, tables, image regions and coordinates. The public directory labels it preview.

The Parse API accepts images only. PDF is rejected before upload; no hidden rasterization or fallback model is used.
Image regions contain original source pixels from returned boxes, not generated images. This is document parsing, not ordinary chat.

Cohere Rerank v4.0 Pro: complete within the verified dedicated deployment scope

Verified Cohere Rerank v4.0 Pro dedicated multilingual document reranking with structured text, top_n, priority and ranked scores.

Reranking only: chat and coding-agent requests are not admitted, regardless of misleading deployment metadata.
The provider truncates document tails at max_tokens_per_doc; HCD does not rewrite source text and discloses that policy. Estimated scores are separate from capability acceptance.

decision-model-preview: Full structured decisions (verified deployment scope)

Classification, yes/no probabilities and ordinal scoring preserve original probabilities, confidence and input usage; no chat text is generated.

Text, object or array state. HCD:1–16 questions, up to255 choices,2–10 score levels,32768 UTF-8 bytes total. Provider input context:65536 tokens.
Rounded probabilities and sums remain unchanged, without normalization. Confidence is not verified accuracy. Currently limited-time free; normal account/Key authorization, metering and single settlement apply.
No chat, streaming, tools, search, cache, Batch, media or persistent jobs. Maximum context not live-tested (high-cost exemption). Generic fixed-panel estimate:0/41 direct coverage, not decision accuracy; value uses this model own input tariff and decision workload.

DeepSeek V4.1 Flash: complete within the verified dedicated deployment scope

Verified DeepSeek Flash text reasoning, tools, streaming/history/cache, native Chat prefix FIM and image input within the exact deployed scope.

Image files use Key-owned authorization and lifecycle handling; compatibility media URLs require HTTPS. Hosted web/file search, native namespace discovery, encrypted reasoning and native background jobs are not exposed.
Official maximum context/output and large-media bounds use guards and offline checks; expensive extremes are not live-tested (high-cost exemption).
Earlier 74.9/75.3 figures were incomplete-panel estimates, not exact-model acceptance scores. Public composite performance/value scores await exact evidence.
Current exact protocol output bounds:393216 tokens for Chat/Responses and4096 for FIM. Former384000 corrected from exact schema; no extreme live test.

DeepSeek V4 Pro: complete in the verified deployment scope

Verified text, structured output, reasoning, functions/client tool discovery and real Codex continuation; native Chat streaming/Prefix, Messages and FIM.

Hosted web search, image/file input and upstream stored/background state are unsupported; the model profile disables web search.
Performance69.3 and value65.7 are estimates:7 exact0813 public metrics and34 estimates in the fixed41-metric panel.
Maximum context/output/concurrency: not live-tested (high-cost exemption); official limits and offline guards apply.
Current exact protocol output bounds:393216 tokens for Chat/Responses and4096 for FIM. Former384000 corrected from exact schema; no extreme live test.

doubao-deepseek-v4-1-flash: verified deployment scope

Text, image and video; four reasoning levels, JSON Schema, functions, streaming and saved history. Actual Codex tools and continuation verified. Native remote MCP and owned Responses; HCD owned-file and controlled web-search bridges.

Context 1,048,576 tokens; input maximum 1,048,576; default/maximum output 393,216 including thought, within the combined context. Compatibility image limit 1,500; native 3,600; each media item 50 MiB, request 64 MiB. Owned inline files: 5 MiB each, 24-hour retention.
Implicit cache is metered. Explicit cache, native hosted search, Flex, Batch and background generation are closed and rejected. HCD MCP search reserves and bills separately. Actual Work client remains unverified. Performance 77.5/value 76.8 are estimates over 41 metrics, 12 direct model sources, no direct offering benchmark. Expensive boundaries checked offline.

doubao-DeepSeek V4 Flash GA (2026-07-31) — Full integration (verified deployment scope)

Native knowledge search has no authorized resource and is closed. Cache with context edits, unregistered native MCP servers, Batch/background/images are closed and rejected before dispatch.
Default low. Deleting a response does not end cache storage liability before TTL. Performance56.6/value58.8 estimated,0/41 direct; high-cost extremes not live-tested.
Account points/M tokens, standard input/output/cache:821918/2465753/27397; flex:410959/1232877/27397. No extra token cache-write fee. Storage4658 points/M-token/hour; search5480 points/call. Key multiplier and per-item rounding apply separately.

doubao-DeepSeek V4 Pro GA (2026-08-13) — Full integration (verified deployment scope)

Verified text, none/low/high/max, JSON Schema, functions, bounded search, full tool history, MCP, implicit/explicit cache and native response resources. Native knowledge search has no authorized resource and is closed. Cache with context edits, unregistered native MCP servers, Batch/background/images are closed and rejected before dispatch. Flex returned upstream ModelLoading and is not verified or open; rejected before dispatch. Its reviewed rate is reference only. Cache continuation must explicitly stay within the original TTL; cross-model references, context edits and native hosted tools are rejected. Missing reasoning breakdown keeps a funded storage upper bound without posting an estimated charge.

Native knowledge search has no authorized resource and is closed. Cache with context edits, unregistered native MCP servers, Batch/background/images are closed and rejected before dispatch.
Default low. Deleting a response does not end cache storage liability before TTL. Performance59.4/value56.2 estimated,7/41 direct; high-cost extremes not live-tested.
Account points/M tokens, standard input/output/cache:2465753/7397260/82192; flex:1232877/3698630/82192. No extra token cache-write fee. Storage4658 points/M-token/hour; search5480 points/call. Key multiplier and per-item rounding apply separately.
Flex returned upstream ModelLoading and is not verified or open; rejected before dispatch. Its reviewed rate is reference only. Cache continuation must explicitly stay within the original TTL; cross-model references, context edits and native hosted tools are rejected. Missing reasoning breakdown keeps a funded storage upper bound without posting an estimated charge.

FLUX.2 Pro: complete within the verified dedicated deployment scope

Verified FLUX.2 Pro dedicated image generation, editing and up to eight reference images. The public directory currently marks it as planned.

PNG/JPEG output and the model-specific prompt enhancement mapping were checked; tuning fields from another model are explicitly rejected.
Image streaming, transparent output, mask editing and steps/guidance controls are outside this exact interface. Image acceptance is neither a quality score nor a chat-route claim.

Google Gemini 3.8 Flash-100: fully reviewed deployed scope

Native GenerateContent, Chat and Interactions support text, image, audio, video and PDF input, thinking, JSON, functions, streaming and continuation. Codex uses the HCD Responses bridge.

Knowledge Studio adds isolated Python (NumPy, SymPy and Matplotlib), URL reading, video-frame inspection and controlled web search. These are HCD bridges; native hosted code, search and AGENTIC video loops are not enabled. Computer use returns proposed actions only.
Combined context: 1,048,576 tokens; no disclosed independent input maximum. Default and maximum output: 65,536 tokens including thinking, subject to remaining context. Explicit cache read/delete/expiry and owned HCD files, background responses and Batch preserve original results. Batch uses standard call prices.
Per million tokens: input 3,000,000 points, cache hits 300,000 and output including thinking 15,000,000. Explicit cache creation uses the input rate; storage costs 2,000,000 points per million token-hours. Supplier promotional credits have not been received and do not promise customer refunds. Performance 76.6 and value 66.1 are estimates: 12 model-wide public metrics and 29 estimated metrics, with no direct performance measurement for this offering. Expensive limits were checked offline.
Work Responses live web search uses an HCD controlled bridge. In auto mode the model decides whether to search, at most once per response; model and search usage are metered separately from original receipts. Authenticated actual Work client and search roundtrip verification remains pending.
Gemini 3.8 Flash-100 supports HCD summary views: auto keeps the provider automatic summary; detailed shows the full available provider summary; concise explicitly shows an HCD bounded extract and truncation status; none hides the public summary. These views do not control provider summary length or add model calls. Thinking tokens retain original receipt billing. Actual remote Work roundtrip verification is pending.

Gemini 3.8 Live-100 · Current deployment fully integrated

Realtime audio/video, camera/images, prebuilt Avatar, function follow-ups, native resume, interruption and bounded web search.

Custom Avatar is unavailable on this deployment and rejected before provider contact. The user explicitly deferred and exempted it from full certification; no application was submitted.
Search and automatic VAD use bounded HCD bridges; provider-native search is not exposed.
Google supplier rates per1M tokens: TEXT/IMAGE-VIDEO/AUDIO input0.75/1/3USD; TEXT/AUDIO/VIDEO output4.5/12/1USD. Thinking uses text-output rates; transcription and retained history remain in original meters. Search has a separate quote.
Performance77.2 and value84.0 are estimates; direct panel coverage0/41. Audio/Avatar quality not benchmarked. Extreme contexts/media: not live-tested (high-cost exemption).

Google Gemini 4 Argon: verified deployed capabilities

Text, image, audio, video and PDF input; three reasoning levels, JSON, function tools and streaming. Codex tools and continuation are verified. HCD knowledge tools provide isolated Python, URL reading, video-frame inspection and separately billed controlled search. Responses builtin web_search uses the bounded HCD compatibility bridge; actual Work client verification remains pending. Preview search and unmapped explicit controls are rejected.

Combined context is 1,048,576 tokens; the independent input limit is not disclosed. Native GenerateContent defaults to and permits 1,048,576 total output tokens through manual owner-bound continuationToken requests; each segment permits 32,768 non-streaming or 65,536 streaming tokens. Responses, Chat and Interactions default and maximum per-call outputs are 32,768 non-streaming / 65,536 streaming, including reasoning within combined context. Explicit cache is unsupported; implicit cache is metered. Owned HCD files, background responses and Batch retain lifecycle controls; Batch uses standard call prices.
Per million tokens: 8,000,000 input points, 400,000 cache-hit points and 40,000,000 output/reasoning points. Full input, output and tool liabilities are reserved against both account and Key before submission; each explicit continuation is separately reserved and settled. Performance 77.2 and value 62.5 are estimates: 6 model-wide public metrics plus 35 estimates in a fixed 41-metric panel, with no direct offering performance measurement. Expensive boundaries were checked offline. Supplier promotional credits are not treated as received.

Gemini Nano Banana 2.1-100: Complete within verified deployment scope

Complete verified image-studio scope; HCD bounds are distinct from manufacturer maxima.

Image generation/editing, streaming and private-reference continuation; 1K/2K/4K, 15 aspect ratios, minimal/medium/high, up to32768 output tokens and native system instructions. One PNG/JPEG result per request.
Up to14 PNG/JPEG/WebP/HEIC/HEIF references:7MiB each,40MiB combined media,64MiB complete request/history. One PDF/UTF8 file up to8MiB/30 PDF pages, or one silent MP4/WebM up to16MiB/60s with crop and fps(0,24].
HCD controlled webpage/source-image search with sources, not Google native search or an image index. The cumulative50 CNY platform risk allowance uses conservative metered supplier upper bounds:30 CNY held per request, concurrency1; new search may stop before actual spending reaches50 CNY. It is not customer credit, free usage or an upstream bill cap.
Search and image components settle once from their original receipts. A successful search may remain charged when later image processing fails. Submitted jobs and result reading continue after the risk stop; ordinary generation/editing works independently.
HCD durable batches support create, disconnect, undispatched restart, status, cancellation and results.1..2 sequential items only:1K/1:1/minimal ormedium/max4096 output/one PNG-JPEG reference/no search. Standard realtime per-item rates apply, without Google native Batch discounts.
Native countTokens is advisory, not generation billing or the sole reservation bound. Input, implicit cache reads, text/thought output and image output are metered separately. Explicit cache/file URI/YouTube/native Batch storage are unavailable; upstream audio/function calling/structured output are unsupported. Original image bytes and provider provenance are preserved.
4K and maximum context/output/14-reference extremes: not live-tested (high-cost exemption); exact model evidence and offline resource boundaries verified. Performance92.5 comes from two public image benchmarks; value66.5 is estimated from standard prices, not route performance testing.

GPT-5.6 Luna: complete within the verified safe deployment scope

The exact model is integrated across its verified safe deployment scope. Expensive limits and unpriced optional hosted features retain explicit restrictions.

Text and image input with text output; reasoning supports none, low, medium, high, xhigh and max, with medium as the default. Structured output, functions, real Codex default tools and a saved-history second turn were verified.
Raw streams and cache metering, background completion after disconnect, owned GET, replay and DELETE were accepted. WebSocket supports a verified two-response response.create chain; mid-response steering is unavailable for this model.
A small valid image was tested for transport, meaning and metering. Bounded images reserve the full documented image-token ceiling. Unknown media and files retain strict liability bounds and may be rejected before dispatch when funds are insufficient.
The long-context tier starts above 272,000 total input tokens, including cache, and prices the whole request. Account and Key calculations for input, cache reads, cache writes and output were checked offline across the boundary. Maximum context/output, live threshold crossing and large media are not live-tested (high-cost exemption).
Pro, multi-agent work, arbitrary remote MCP, indexed search, containers, hosted tools and image generation remain closed where price or lifecycle bounds are unverified. Parameters are not silently discarded. Model audio/video input and direct non-text output are unsupported.
Current performance 64.2 and value 65.8 derive from public evaluations and coherent pricing, not an HCD live quality score.

GPT-5.6 Sol: complete within the verified safe deployment scope

Completion covers the exact model and verified safe deployment. Optional hosted capabilities and expensive boundaries retain explicit limits. Reservations, settled charges and unresolved liability stay separate.

Text and image input with text output; reasoning supports none, low, medium, high, xhigh and max, with medium as the default. Structured output, native Chat functions with explicit none, Responses functions, real Codex default tools and two-turn saved history were verified.
Raw streams, cache writes and reads, background completion after delivery disconnect, owned GET, stream replay and DELETE were accepted. WebSocket scope is text, low/default, 64 output tokens, one ordinary response.create continuation and two responses. This model does not support response.steer. The connector continues to use HTTP/SSE.
A small valid image was tested for transport, meaning and metering. Bounded images reserve the full documented image-token ceiling. Unknown media and files retain strict full-context liability or are rejected before provider contact when funds are insufficient.
The long-context tier starts above 272,000 total input tokens, including cache, and applies to the whole request across input, cache reads, cache writes and output. Account and Key math was checked offline on both sides of the threshold. Maximum context/output, live threshold crossing and large media are not live-tested (high-cost exemption).
Pro, multi-agent, arbitrary remote MCP, indexed file search, containers, hosted Shell/Skills, hosted image generation and unverified sampling parameters remain closed; no parameter is silently discarded. Model audio/video input and direct non-text output are unsupported.
Existing performance 88.2 and value 70.5 use public evaluation evidence and coherent reviewed pricing, not HCD live quality scores.

GPT-5.6 Terra: complete within the verified safe deployment scope

The exact model is integrated across its verified safe deployment scope. Expensive limits and unpriced optional hosted features retain explicit restrictions.

Text and image input with text output; reasoning supports none, low, medium, high, xhigh and max, with medium as the default. Structured output, functions, real Codex default tools and a saved-history second turn were verified.
Raw streams and cache metering, background completion after disconnect, owned GET, replay and DELETE were accepted. WebSocket supports a verified two-response response.create chain; mid-response steering remains closed pending exact-model verification.
A small valid image was tested for transport, meaning and metering. Bounded images reserve the full documented image-token ceiling. Unknown media and files retain strict liability bounds and may be rejected before dispatch when funds are insufficient.
The long-context tier starts above 272,000 total input tokens, including cache, and prices the whole request. Account and Key calculations for input, cache reads, cache writes and output were checked offline across the boundary. Maximum context/output, live threshold crossing and large media are not live-tested (high-cost exemption).
Pro, multi-agent work, arbitrary remote MCP, indexed search, containers, hosted tools and image generation remain closed where price or lifecycle bounds are unverified. Parameters are not silently discarded. Model audio/video input and direct non-text output are unsupported.
Current performance 73.8 and value 62.9 derive from public evaluations and coherent pricing, not an HCD live quality score.

GPT-6 Astra: complete within the verified, safely available deployment scope

Completion covers the verified deployment scope, not every optional hosted capability. Known prices and provable quantity bounds govern availability. Holds and settled charges remain separate; authorization and budget gates remain active.

Verified text, low/medium/high/xhigh/max reasoning, structured output, Responses functions/namespaces/parallel and async tools, tool search, client computer/patch protocols, streaming, saved and encrypted history, cache, compaction and background resource lifecycle. Actual Codex default tools and saved-history continuation were independently tested.
PNG and bounded simple single-frame WebP/non-animated GIF were verified. The latter use low detail, dimensions up to 512 pixels and files up to 1 MiB. JPEG transport and metering passed, but a red fixture was described as Purple; that semantic case failed. Unknown media and remote URLs retain strict holds or rejection.
WebSocket scope is text, low reasoning, 64 output tokens, one queued steer and an automatic second response. The original response completes normally; immediate interruption and unverified tool/history combinations are not enabled.
The deployment rejects temperature on native Responses, and logit_bias/function tools on native Chat. Local guards remain. Other sampling controls and max_tokens are not verified; use max_completion_tokens for Chat and max_output_tokens for Responses.
file_search index size and retention bounds remain unresolved. Code Interpreter/container pricing has conflicting minute versus Session units. Pro/multi_agent aggregate quantities and costs lack proven bounds. These features remain restricted or disabled.
Hosted remote MCP was tested only with an owned fixed HTTPS echo, same-Key/project approval and two metered responses. Its temporary entry is closed. Arbitrary remote servers, variable results, credentials and broader lifecycle are not enabled; local MCP tests are not substitutes.
Batch, Flex and Fast/Priority are not offered by this deployment. Audio/video input and direct media output are outside its verified scope. Unavailable hosted image generation, Shell/Skills and other optional tools remain gated.
The 272,000-token tier uses total input including cache. An actual 272,022-input-token long-tier receipt was verified. Maximum context/output and other expensive boundaries retain offline checks and explicit high-cost live-test exemptions, not claims of maximum-size live generation.

GPT-6 Luna: complete within the verified, safely available deployment scope

Completion covers the verified safe deployment. Optional upstream features remain limited where bounds are unproved. Reservation and settlement stay separate, with exact-model authorization and cost gates.

Verified text, default medium and none/max, structured output, original streams, Chat functions with none, Responses async functions, a small image semantic case, actual Codex default tools and in-memory MCP across two turns, cache usage, background disconnect/completion/retrieve/stream replay/delete, and bounded WebSocket steering with two responses.
Reasoning choices are none, low, medium, high, xhigh and max, with medium as default. Chat functions require explicit none; max uses Responses.
WebSocket is limited to native text, low reasoning, default tier, 64 output tokens, one queued steer and two responses. The connector uses HTTP/SSE.
Maximum context/output, long-context threshold, large media and high-output compaction use official limits and offline checks: not live-tested (high-cost exemption). A small image was tested, but unknown media retains full-context reservation; no low-cost image bound is claimed.
Pro, multi-agent, automatic compaction, arbitrary remote MCP, indexed file search, containers and hosted Shell/Skills stay restricted where safe cost or lifecycle bounds are unproved. Model audio/video input and direct non-text model output are unsupported. Unverified sampling parameters are explicitly rejected.
Some exact-model public benchmarks exist, but the fixed 41-cell scoring panel has not passed a like-for-like review. Composite performance and value scores remain pending; no other model's estimate is reused.

GPT-6 Sol: complete within the verified, safely available deployment scope

Completion covers the verified safe deployment, not all optional official capabilities. Reservation bounds and actual settlement remain separate; exact-model authorization and cost gates remain active.

Verified text, default medium and none/max parameters, structured output, original streams, Chat function calling with none, Responses async functions, actual Codex default tools and in-memory MCP with two persisted turns, cache usage, and background disconnect/completion/replay/retrieve/delete.
Reasoning choices are none, low, medium, high, xhigh and max, with medium as the model default. Chat functions require explicit none; max uses Responses.
WebSocket availability is native text with low reasoning, default tier, 64 output tokens, one queued steer and two responses. The connector continues to use HTTP/SSE.
Maximum context/output, long-context thresholds, large media and high-output compaction use exact official limits and offline boundary checks: not live-tested (high-cost exemption). Unknown media retains full-context reservation; no low-cost or live image-understanding claim is made.
Conflicting official sampling descriptions remains unresolved, so those parameters are rejected. Pro, multi-agent, automatic compaction, arbitrary remote MCP, indexed file search, containers and hosted Shell/Skills remain restricted where safe cost or lifecycle bounds are unproved. Model audio/video inputs and direct non-text model outputs are unsupported.
Composite performance and value scores are marked estimated: 13 of 41 panel cells have exact-model public observations and 28 use same-metric peer estimates. These are not HCD deployment tests. Value uses reviewed public account-point rates, not actual charges. Rankings against GPT-5.6 Sol and other models apply only to this fixed panel; overlapping estimate ranges do not establish general capability order. Image semantics and dedicated tool-search/computer/apply-patch semantic cases were not separately live-tested.

GPT-6.1 Sol: complete within the verified deployment scope

The exact model is integrated across its currently verified scope. Expensive boundaries and unpriced hosted capabilities retain explicit restrictions.

Text and image input, text output. Five reasoning efforts: low, medium, high, xhigh and max; medium is default. none/minimal and native Chat tool calling are unsupported; use Responses for tools.
Strict structured output, functions and async options, streams and cache metering, real Codex default tools, an in-memory MCP round trip and a persisted second turn were verified.
Background completion after delivery disconnect, owned GET, stream replay and DELETE were verified. Narrow WebSocket allows one ordinary response.create continuation; steer remains pending and closed. Codex uses HTTP/SSE.
Points use the user-approved exact-model official tariff as an estimate. Supplier formal rates remain unconfirmed. Estimated pricing is used for reservation and account/Key service settlement, not confirmed supplier charges.
The whole request switches to long-context pricing above 272,000 total input tokens including cache. Maximum context/output, threshold crossing and large media are not live-tested (high-cost exemption); official limits and offline boundaries apply.
Unknown media/files retain full-context liability bounds and are rejected before dispatch when budgets are insufficient. Unpriced Pro, multi-agent, automatic compaction and hosted resources stay closed. Performance and value await exact numerical evidence.

GPT Audio: complete within the verified dedicated deployment scope

Verified dedicated audio inference with text/audio input and text/audio output, streaming and function calls. The public directory currently marks this entry as planned; access depends on the actual entry permissions.

This is acceptance of the exact audio interface, not a general Codex conversation-model claim. Image I/O, realtime sockets, structured-output schemas and discounted prompt cache are outside the verified scope.
Signed input limits bound resources; unsupported parameters are rejected before submission. Capability acceptance is not a quality benchmark, and directory scores retain their stated evidence grade.

GPT Audio 1.5: complete within the verified dedicated deployment scope

Verified GPT Audio 1.5 dedicated audio inference: text/audio input and text/audio output, streaming and function calls. The public directory currently marks it as planned.

Acceptance covers this exact audio interface; it does not activate a general Codex chat route. Images, realtime sockets, structured-output schemas and discounted cache pricing are outside scope.
Signed input contracts bound resources and reject unsupported parameters before submission. Directory scores are separate from this capability acceptance.

GPT Audio Mini: complete within the verified dedicated deployment scope

Verified GPT Audio Mini dedicated text/audio I/O, streaming and function calls. The public directory currently marks it as planned.

This is not a general Codex chat integration. Image I/O, realtime sockets and structured-output schemas are outside scope.
Cache hints and native usage reporting were verified, but live calls reported zero cached tokens; no actual hit or discount was proven. Unpublished discounted rates are not guessed.
Functional acceptance is not a quality benchmark; estimated directory scores remain estimates.

GPT Live Transcribe: complete within the verified dedicated deployment scope

Verified dedicated live audio transcription with context hints and streaming. The public directory currently marks it as planned.

This is a transcription entry, not general chat, function tools, a file-transcription endpoint or audio generation.
Signed contracts bound inputs and reject unsupported fields; capability acceptance is not a transcription-quality score.

GPT Realtime: complete within the verified dedicated deployment scope

Verified dedicated realtime text/audio/image input, text/audio output, streaming, functions and prompt caching. The public directory currently marks it as planned.

The entry uses HCD browser audio transport and a WebSocket proxy; it does not establish full WebSocket support for ordinary Codex chat clients.
Generated image/video output, structured-output schemas, reasoning-effort control, external SIP and hosted remote MCP are not offered. Functional acceptance is not a quality score.

GPT Realtime 1.5: complete within the verified dedicated deployment scope

Verified GPT Realtime 1.5 dedicated realtime text/audio/image input, text/audio output, streaming, functions and caching. The public directory currently marks it as planned.

Acceptance covers browser audio transport and its WebSocket proxy, not full WebSocket capability in general Codex clients.
Generated image/video output, structured-output schemas, reasoning-effort control, external SIP and hosted remote MCP are unavailable; directory scores are separate from this acceptance.

GPT Realtime Mini: complete within the verified dedicated deployment scope

Verified GPT Realtime Mini dedicated realtime text/audio/image input, text/audio output, streaming, functions and caching. The public directory currently marks it as planned.

Scope is browser audio transport and its WebSocket proxy, not full WebSocket support for ordinary Codex clients.
Generated image/video output, structured-output schemas, reasoning-effort control, external SIP and hosted remote MCP are unavailable. Capability acceptance and estimated scores are separate.

Grok 4.7 — Full verified deployment scope

Native Chat, automatic-reasoning Responses and a four-effort Codex Chat bridge support text/images, JSON, tools and saved-thread history.

low, medium, high and xhigh passed live; high is the default and none is unsupported. Codex uses the Chat bridge; native Responses uses automatic reasoning.
Output includes reasoning tokens. Input including cache reads above200000 tokens applies the2x long-context tariff to the whole request. Cache hints are explicitly not forwarded; cache hits are not promised.
Actual full Codex tools, MCP echo and saved-thread continuation passed. Performance70.3 and value61.4 are estimates with0/41 direct coverage. Expensive context/media limits were not live-tested.
No provider-hosted Web/X/code tools or encrypted state claimed for this deployment. Google-100 has its own reviewed scope and prices.
xAI states No text output limit: no separate text-output cap. The500000 combined context still applies; HCD uses a finite supported ceiling and upstream actual-token handling, never reduces the default using fee estimates or claims a second independent500000 output budget.

Google Grok 4.7-100 — Fully reviewed deployed scope

Native Responses and Chat: text/images, four reasoning levels, JSON, complete tools and encrypted continuation.

low, medium, high and xhigh reviewed; high is the default and reasoning cannot be disabled. Adjustable native Google Responses was live verified.
Actual connector-generated Codex configuration, complete tools, MCP echo and same saved-thread continuation passed. Encrypted reasoning was replayed unchanged; cache keys and real cache hits were verified.
Performance70.3 and value61.4 are estimates, direct coverage0/41. Expensive context/media extremes were checked offline. Provider-hosted Web/X/code tools, Batch, upstream files, reasoning summaries and cache retention are not exposed.
Combined context524288 tokens; independent output maximum not disclosed. HCD context_bound support and default are constrained by remaining context, not a published independent output maximum.

MAI Thinking 1: complete within the verified dedicated deployment scope

Verified MAI Thinking 1 native text reasoning, function tools, history continuation, streaming/nonstreaming and token usage for a dedicated text-tool conversation.

The Responses bridge validates a complete result before replay; incremental Responses delivery is not implemented. Native API streaming preserves original events.
The directory labels it preview. Client support follows the verified entry and does not imply that every Codex function is exposed.

Mistral Medium 3.5: complete within the verified dedicated deployment scope

Verified Mistral Medium 3.5 native text/image input, text chat, JSON objects, streaming, multiple candidates and sampling through a dedicated text/code-tool entry.

The exact hosted offering does not advertise tool calling; HCD rejects tool fields before generation and does not present it as an ordinary Codex tool agent.
Explicit temperature=null or top_p=null is rejected locally after upstream rejection; omission uses defaults. The directory labels it preview, and estimated scores are separate.

Mistral OCR 4: complete within the verified dedicated deployment scope

Verified Mistral OCR 4.0 dedicated image/PDF recognition, text/table/image extraction, layout, page selection and multilingual OCR. This is a document entry, not a chat model.

The exact deployment accepts token, word and page confidence granularity; upstream rejects block granularity with 422, so it is rejected before submission.
The public directory labels it preview. Capability acceptance and performance/value score evidence remain separate.

通用多模态向量: complete within the verified dedicated deployment scope

Verified Multimodal Embedding v1 dedicated independent text, image and video vectors with ordered mixed batches. The public directory currently marks it as planned.

At most 20 inputs per batch, each producing a fixed 1024-dimensional vector. The provider processes at most 512 tokens per text item and truncates longer text; HCD preserves the submitted text and discloses this behavior.
Dense output, image URL/local input and relevant parameter forwarding are supported. No measured retrieval gain, modality fusion, variable dimensions, streaming or async Batch is claimed.
Performance 62.3 and value 48.1 remain low-confidence estimates, not direct exact-model evaluation. Media reservations and settled charges are separate.

qwen-DeepSeek-V4-Flash-0731：Verified integration (currently open scope)

Full integration within the verified open deployment scope: text, none/low/high/max, structured output, functions, complete default Codex tools, pure in-memory MCP, explicit saved-thread continuation and implicit cache. Hosted search/extraction/interpreter and native storage/response-ID/conversation/background are closed and explicitly rejected before provider dispatch.

Closed because aggregate agent iteration/charge bounds are unverified. max_output_tokens limits reply only; undocumented max_tool_calls is not a hard cap. The separate393216 reasoning-token maximum is reserved; high-cost extremes are not live-tested.
Original search receipts and historical charges are preserved; hosted search is currently closed. Own prices and scores unchanged.
Current official reply/reasoning total shares393216; reservation uses that shared ceiling, preserving separate raw usage and once-only settlement.

qwen-DeepSeek-V4-Pro · Verified open deployment scope

Full integration within the verified open deployment scope: text, none/high/max, structured output, functions, complete default Codex tools, pure in-memory MCP, explicit saved-thread continuation and implicit cache. Hosted search/extraction/interpreter and native storage/response-ID/conversation/background are closed and explicitly rejected before provider dispatch.

Performance 61.3, value 56.9: estimated fixed panel with 7 model-wide public observations and 34 estimates. Own standard tariff 3287671/6575342/cache273973 points per million tokens; no peak/off-peak tariff. Provider alias has no declared immutable snapshot. Separate Responses thinking reservation uses a conservative deployed output upper bound; expensive limits not live-tested. Supplier invoice cash remains unknown.
Current official reply/reasoning total shares393216; reservation uses that shared ceiling, preserving separate raw usage and once-only settlement.

qwen-DeepSeek-V4-Pro-0813 · Verified open deployment scope

Closed because aggregate agent iteration/charge bounds are unverified. max_output_tokens limits reply only; undocumented max_tool_calls is not a hard cap. The separate393216 reasoning-token maximum is reserved; high-cost extremes are not live-tested.
Performance71.7/value67.7 estimated;7 exact public model metrics and34 estimates in41 fixed panel. Value uses this offering's reviewed off-peak base workload; cache separate. Off-peak22:00–08:00 UTC+8; peak token rates x2. Supplier invoice cash unknown.
Current official reply/reasoning total shares393216; reservation uses that shared ceiling, preserving separate raw usage and once-only settlement.

Qwen-Image-2.1-Pro：complete within the verified deployment scope

Image-studio-only model; completion covers verified generation, editing, parameters, metering and native task lifecycle.

Unified generation/editing, up to10 reference images and1–6 outputs; transparent images, seed, prompt enhancement, Direct/Agent, enable_thinking and watermark. Agent is generation-only.
Static PNG/JPEG/WebP/BMP/TIFF/GIF inputs; native PNG output, with HCD JPEG/WebP conversion. Masks guide reference editing; HCD completion events are not native partial-image streaming.
Native async tasks return result_ref and continue after client disconnect. Original usage settles once; same-Key original-result reads are free. Provider results expire after24 hours; save returned images.
HCD bounds:18,000 prompt characters,10MB per image,64MiB aggregate,40M input pixels,8000-pixel sides and10:1 input ratio. Output area512²–2048² and ratio1:8–8:1. Six-output/ten-reference extremes use offline checks and high-cost live exemptions.
No ordinary Chat, function tools, web search, context cache or native Batch inference. Performance is a same-family public-data estimate, separate from functional acceptance.

Qwen3-VL-Rerank: complete within the verified dedicated deployment scope

Verified Qwen3 VL Rerank synchronous text, image, video and mixed-input reranking. The public directory currently marks it as planned.

Image queries, top_n, instructions, returned documents and native responses are supported. Video fps is strictly 0 < fps <= 1; exact deployment rejects zero and HCD rejects it before provider dispatch.
Official 4,000 query and 8,000 per-item token descriptions conflict. Exact limits remain upstream-enforced pending clarification; no invented truncation is applied.
493154 points per million tokens for video-containing requests is an approved liability estimate, not a published supplier tariff or verified supplier cost. This synchronous entry has no streaming, Batch or file jobs.

千问3.7-通用文本重排序: complete within the verified dedicated deployment scope

Verified Qwen3.7 Text Rerank synchronous text reranking, instructions, top_n, all results and large candidate sets. The public directory currently marks it as planned.

Each document is limited to 50,000 characters; over-limit input is rejected before submission without truncation.
This dedicated reranking entry has no image/video, streaming, hosted tools, uploaded files or async jobs. Parameters from other models are not substituted. Estimated scores are not quality results from this acceptance.

현재 경계

현재 제공 경계

Google 이외 모델 32개를 추가했습니다: 텍스트 27개, Mistral Document AI, 임베딩 3개, Sora 2. 연결 도구 1.0.86.0으로 업데이트하고 각 플러그인과 API 키 권한을 켜세요. 텍스트 API는 34개, Codex 메뉴는 도구를 지원하는 32개 모델입니다. Phi와 Llama는 API 질의용입니다. 문서는 최대 30 MB·30페이지입니다. 지식 검색은 선택한 로컬 문서를 인용하며 색인을 재사용합니다. 비디오는 4/8/12초·720p, 참조 이미지와 완성 영상 리믹스를 지원합니다. 영상을 신속히 다운로드하세요. 완료된 사용량만 청구하며 로컬 캐시 읽기는 무료입니다. Google과 model-router는 제외됩니다. 설비 식별자, 프로토콜, 운영 데이터, 에이전트와 제어는 고객이 승인한 로컬, 엣지 또는 전용 환경에 유지됩니다. 공개 웹사이트는 서비스 소개용이며 클라우드 설비 제어 플랫폼이 아닙니다.

프로젝트 문의 로그인