Comparison cohort: vals-biomysterybench:1:overall
Metric: accuracy
Scale: 0-100
Participants: 13
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-22. Harness: Terminus 2. Latency in seconds: 739.252. Cost per task in USD: 2.972555. Source ID: vals-ai. Evidence capture ID: vals-biomysterybench-2026-09-23-1da8cd6684f0. Source payload hash: 1da8cd6684f0e235fa297f17a076add04c169bb4739ba048aa3300d9e5028eaa. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

MysteryMechanism

Overall

1.GPT-6 Astra53.15%

2.Claude Opus 5.549.55%

3.Claude Fable 5.147.75%

4.Claude Opus 537.39%

5.Gemini 3.8 Flash36.49%

6.GPT-5.6 Sol33.33%

7.Grok 4.630.63%

8.Grok 4.725.23%

9.Qwen 3.8 Max 090223.87%

10.GLM-5.322.97%

11.DeepSeek-V4.1-Flash21.17%

12.GPT-6 Luna19.37%

13.GPT-5.6 Luna14.41%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-mysterymechanism:1:overall
Metric: accuracy
Scale: 0-100
Participants: 13
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 1178.454. Cost per task in USD: 4.059711. Source ID: vals-ai. Evidence capture ID: vals-mysterymechanism-2026-09-23-63be0832dc5b. Source payload hash: 63be0832dc5bfc3769dae7055df0a6f6c1b46b0da25663f9022faff971e28703. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

Coding

Code Migration

Overall

1.GPT-6 Astra67.74%

2.Claude Opus 5.566.65%

3.Claude Opus 557.47%

4.GPT-6 Sol57.2%

5.Claude Fable 555.07%

6.Claude Fable 5.154.61%

7.GPT-5.6 Sol52.92%

8.GPT-5.6 Terra47.8%

9.Claude Opus 4.847.25%

10.DeepSeek-V4.1-Flash45.62%

11.GPT-5.545.16%

12.Grok 4.744.82%

13.Grok 4.644.57%

14.GPT-5.6 Luna44.55%

15.Claude Sonnet 544.39%

16.GLM-5.344.22%

17.Claude Opus 4.743.88%

18.GPT-6 Luna42.55%

19.Claude Sonnet 4.639.89%

20.GLM-5.237.87%

21.Grok 4.536.6%

22.Gemini 3.8 Flash36.55%

23.GPT-5.434.98%

24.Gemini 3.7 Flash34.8%

25.Gemini 3.6 Flash30.93%

26.Kimi K2.627.77%

27.Gemini 3.5 Flash26.75%

28.DeepSeek-V4-Pro26.2%

29.GLM-5.125.77%

30.Kimi K2.7 Code25.39%

31.Qwen 3.8 Max 090223.96%

32.MiMo-V2.5-Pro21.56%

33.GLM-5.3-Flash20.52%

34.MiniMax M319.93%

35.Gemini 3.1 Pro Preview17.31%

36.Kimi K316.1%

37.GPT-5.4 nano14.47%

38.MiMo-V2.514.23%

39.Qwen 3.8 27B14.16%

40.Inkling Small13.68%

41.Qwen 3.7 Max13.07%

42.GPT-5.4 mini12.94%

43.Qwen 3.7 Plus12.86%

44.Inkling11.79%

45.MiniMax M2.78.69%

46.Kimi K2.56.96%

47.Grok 4.36.8%

48.Gemini 3 Flash Preview6.39%

49.Gemini 3.5 Flash-Lite6.15%

50.Mistral Medium 3.55.13%

51.Nemotron 3 Ultra4.91%

52.Gemini 3.1 Flash-Lite4.61%

53.Nemotron 3.5 Lightning2.95%

54.Grok 4.20 Reasoning0.31%

55.Ling 3.0 Flash0%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-code-migration:1:overall
Metric: accuracy
Scale: 0-100
Participants: 55
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 11925.532. Cost per task in USD: 69.480849. Source ID: vals-ai. Evidence capture ID: vals-code-migration-2026-09-23-c7fade0e4383. Source payload hash: c7fade0e4383b27bca9fb2e591abb1b60950f35fca099c1ae51cad7c14fb7292. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

IOI

Overall

1.GPT-6 Astra100%

2.Claude Opus 5.595.06%

3.GPT-5.6 Sol91.17%

4.Claude Fable 5.190.78%

5.GPT-5.6 Terra87.61%

6.Claude Opus 584.33%

7.GPT-6 Sol82.61%

8.Qwen 3.8 Max 090268.89%

9.GLM-5.368.44%

10.Gemini 3.7 Flash67.83%

11.GPT-5.6 Luna61.78%

12.Grok 4.757.72%

13.Gemini 3.8 Flash56.94%

14.GPT-6 Luna55.56%

15.GPT-5.3-Codex53.83%

16.GLM-5.3-Flash52.5%

17.Kimi K348.94%

18.Grok 4.647.61%

19.Claude Sonnet 545%

20.Grok 4.540.56%

21.DeepSeek-V4.1-Flash40.28%

22.Qwen 3.8 27B39.06%

23.Gemini 3.6 Flash35.06%

24.Inkling14.94%

25.Inkling Small9.33%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-ioi:2:overall
Metric: accuracy
Scale: 0-100
Participants: 25
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-22. Latency in seconds: 1035.191. Cost per task in USD: 4.382385. Source ID: vals-ai. Evidence capture ID: vals-ioi-2026-09-23-81487e138796. Source payload hash: 81487e1387966f49c2cce85653523e20447fd6ca8cee8e30372da5b934f8fa0c. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

ProgramBench

Overall

1.Claude Opus 5.518.5%

2.Claude Fable 5.17%

3.GPT-6 Astra5.5%

4.Claude Opus 53%

5.Claude Fable 52%

5.GPT-6 Sol2%

5.Kimi K32%

8.GLM-5.31.5%

8.GPT-5.6 Sol1.5%

10.Claude Opus 4.81%

10.Gemini 3.8 Flash1%

12.Claude Sonnet 4.60.5%

12.GLM-5.20.5%

12.GPT-5.50.5%

12.GPT-5.6 Terra0.5%

12.GPT-6 Luna0.5%

12.Grok 4.70.5%

12.Inkling Small0.5%

19.Claude Opus 4.70%

19.Claude Sonnet 50%

19.DeepSeek-V4-Pro0%

19.Gemini 3 Flash Preview0%

19.Gemini 3.1 Flash-Lite0%

19.Gemini 3.1 Pro Preview0%

19.Gemini 3.5 Flash0%

19.Gemini 3.5 Flash-Lite0%

19.Gemini 3.6 Flash0%

19.Gemini 3.7 Flash0%

19.GLM-5.10%

19.GLM-5.3-Flash0%

19.GPT-5.40%

19.GPT-5.4 mini0%

19.GPT-5.6 Luna0%

19.Grok 4.30%

19.Grok 4.50%

19.Inkling0%

19.Kimi K2.60%

19.Kimi K2.7 Code0%

19.MiniMax M2.70%

19.Nemotron 3 Ultra0%

19.Qwen 3.8 27B0%

19.Qwen 3.8 Max 09020%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-programbench:1:overall
Metric: accuracy
Scale: 0-100
Participants: 42
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-22. Latency in seconds: 8660.211. Cost per task in USD: 42.65866. Source ID: vals-ai. Evidence capture ID: vals-programbench-2026-09-23-083c733e7512. Source payload hash: 083c733e7512bc651ac31bb15b24fc2b0e6881e24a20748e847265e5556a012a. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

Finance

EMB

Overall

1.Claude Fable 5.176.67%

2.Claude Opus 5.575.94%

3.Claude Fable 573.67%

4.Claude Opus 573.56%

5.GPT-5.6 Sol72.34%

6.Gemini 3.8 Flash72.2%

7.GPT-6 Astra71.7%

8.GPT-6 Sol71.53%

9.Gemini 3.7 Flash71.33%

10.Claude Opus 4.869.37%

11.GPT-6 Luna68.52%

12.GPT-5.6 Luna67.12%

13.Grok 4.766.99%

14.Kimi K366.4%

15.Claude Sonnet 566.32%

16.GPT-5.6 Terra66.21%

17.Gemini 3.6 Flash65.41%

18.GPT-5.564.54%

19.Claude Opus 4.763.73%

20.Gemini 3.5 Flash63.55%

21.Grok 4.662.73%

22.GLM-5.261.53%

23.Claude Sonnet 4.660.15%

24.Qwen 3.8 Max 090260.07%

25.Qwen 3.8 27B59.66%

26.Kimi K2.657.86%

27.DeepSeek-V4.1-Flash57.21%

28.Qwen 3.7 Max57.04%

29.GLM-5.356.34%

30.GLM-5.3-Flash55.93%

31.MiMo-V2.5-Pro55.23%

32.MiMo-V2.555.09%

33.Grok 4.552.91%

34.Gemini 3.1 Pro Preview52.62%

35.DeepSeek-V4-Pro51.62%

36.Qwen 3.7 Plus49.35%

37.MiniMax M347.76%

38.GPT-5.4 mini45.43%

39.GPT-5.4 nano44.75%

40.Gemini 3.5 Flash-Lite43.23%

41.Inkling38.36%

42.Inkling Small31.76%

43.Nemotron 3 Ultra31.74%

44.Kimi K2.528.47%

45.MiniMax M2.728.45%

46.Gemini 3 Flash Preview26.02%

47.Ling 3.0 Flash25.01%

48.Grok 4.318.39%

49.Nemotron 3.5 Lightning16.99%

50.Grok 4.20 Reasoning11.74%

51.Mistral Medium 3.511.19%

52.Gemini 3.1 Flash-Lite8.63%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-emb:1:overall
Metric: accuracy
Scale: 0-100
Participants: 52
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 2106.434. Cost per task in USD: 9.330415. Source ID: vals-ai. Evidence capture ID: vals-emb-2026-09-23-0e761c761c4e. Source payload hash: 0e761c761c4e6aa3d255a9a37e58f22915cd69f36faede07660f9a663ff4a6dc. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

Finance Agent (v2)

Overall

1.Gemini 3.8 Flash61.44%

2.Gemini 3.7 Flash59.04%

3.Claude Fable 5.158.88%

4.Claude Opus 558.63%

5.Claude Opus 5.558.59%

6.Gemini 3.5 Flash57.86%

7.GLM-5.3-Flash57.85%

8.Claude Fable 556.31%

9.Gemini 3.6 Flash56.3%

10.GLM-5.355.84%

11.GPT-5.6 Luna55.04%

12.GPT-5.6 Terra54.44%

13.Kimi K354.36%

14.Claude Opus 4.853.92%

15.Claude Sonnet 553.91%

16.GPT-5.6 Sol53.76%

17.Grok 4.653.68%

18.GPT-6 Astra53.54%

19.DeepSeek-V4.1-Flash53.48%

20.Grok 4.752.25%

21.GPT-5.551.76%

22.Claude Opus 4.751.51%

23.Claude Sonnet 4.651.04%

24.Qwen 3.8 Max 090250.59%

25.GPT-6 Luna49.87%

26.GLM-5.249.7%

27.GPT-6 Sol49.05%

28.Qwen 3.8 27B48.55%

29.Grok 4.548.35%

30.MiniMax M348.27%

31.Qwen 3.7 Max47.78%

32.Gemini 3.5 Flash-Lite47.44%

33.Inkling46.6%

34.GPT-5.4 mini45.36%

35.Kimi K2.644.9%

36.GLM-5.144.79%

37.DeepSeek-V4-Pro44.08%

38.Gemini 3.1 Pro Preview42.98%

39.Gemini 3 Flash Preview42.55%

40.MiMo-V2.5-Pro41.5%

41.Inkling Small41.26%

42.Qwen 3.7 Plus38.22%

43.GPT-5.4 nano38.22%

44.Grok 4.337.73%

45.Nemotron 3 Ultra37.67%

46.MiMo-V2.536.73%

47.Kimi K2.535.79%

48.Mistral Medium 3.532.1%

49.Ling 3.0 Flash30.31%

50.Gemini 3.1 Flash-Lite29.99%

51.Grok 4.20 Reasoning28.49%

52.MiniMax M2.727.89%

53.Nemotron 3.5 Lightning18.5%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-finance-agent-v2:2:overall
Metric: accuracy
Scale: 0-100
Participants: 53
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 1987.81. Cost per task in USD: 9.025984. Source ID: vals-ai. Evidence capture ID: vals-fabv2-2026-09-23-8253d7a09218. Source payload hash: 8253d7a092188677e18f64165f4f338e5b99be804498c7d5476f877cd246914b. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

Legal

Harvey's Legal Agent Benchmark

Overall · Task fully resolved

1.Grok 4.615.83%

2.Grok 4.512.92%

3.Grok 4.712.5%

4.Claude Fable 511.25%

4.Qwen 3.8 27B11.25%

6.Kimi K310.83%

7.Qwen 3.8 Max 090210.42%

8.Gemini 3.8 Flash10%

9.Claude Opus 4.89.58%

10.Gemini 3.7 Flash8.75%

11.GLM-5.38.33%

12.GLM-5.27.08%

13.Claude Fable 5.16.67%

13.Claude Opus 4.76.67%

13.Claude Opus 56.67%

13.DeepSeek-V4.1-Flash6.67%

13.GLM-5.3-Flash6.67%

18.GPT-6 Astra5.42%

19.Claude Sonnet 4.65%

19.Claude Sonnet 55%

21.MiniMax M34.17%

22.Claude Opus 5.53.75%

22.DeepSeek-V4-Pro3.75%

22.GPT-5.53.75%

25.Gemini 3.6 Flash3.33%

26.GPT-6 Luna2.92%

27.Gemini 3.5 Flash2.5%

27.GPT-5.6 Sol2.5%

29.Inkling2.08%

29.MiMo-V2.5-Pro2.08%

31.GPT-6 Sol1.67%

31.Inkling Small1.67%

31.Kimi K2.61.67%

31.MiMo-V2.51.67%

31.Qwen 3.7 Max1.67%

36.GPT-5.6 Luna1.25%

36.Ling 3.0 Flash1.25%

38.GPT-5.6 Terra0.83%

39.Grok 4.30.42%

39.Mistral Medium 3.50.42%

39.Nemotron 3 Ultra0.42%

42.Gemini 3 Flash Preview0%

42.Gemini 3.1 Flash-Lite0%

42.Gemini 3.1 Pro Preview0%

42.Gemini 3.5 Flash-Lite0%

42.GLM-5.10%

42.GPT-5.40%

42.GPT-5.4 mini0%

42.GPT-5.4 nano0%

42.Grok 4.20 Reasoning0%

42.Kimi K2.50%

42.MiniMax M2.70%

42.Nemotron 3.5 Lightning0%

42.Qwen 3.7 Plus0%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-legal-agent-benchmark:1:overall
Metric: task resolution rate
Scale: 0-100
Participants: 54
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-22. Latency in seconds: 2939.571. Cost per task in USD: 17.569061. Source ID: vals-ai. Evidence capture ID: vals-hlab-2026-09-23-507154b1cf0d. Source payload hash: 507154b1cf0dcf67b4d36cb1325228e600b8f727addbed9e862cdc0b61b3af0c. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

Legal Research Bench

Overall · All-pass

1.Claude Fable 5.155.29%

1.Claude Opus 555.29%

3.Claude Opus 5.550.48%

4.Claude Fable 549.52%

5.GLM-5.349.04%

6.GPT-5.6 Sol48.08%

6.Grok 4.648.08%

8.Qwen 3.8 Max 090247.6%

9.Grok 4.747.12%

10.GLM-5.3-Flash45.19%

11.Kimi K344.23%

12.Claude Opus 4.843.75%

13.Claude Sonnet 541.83%

14.DeepSeek-V4.1-Flash41.35%

14.GPT-5.6 Terra41.35%

16.GPT-5.540.39%

17.GPT-6 Astra39.42%

18.Gemini 3.8 Flash38.94%

19.Claude Opus 4.738.46%

19.Claude Sonnet 4.638.46%

21.Grok 4.537.98%

22.GPT-5.6 Luna36.54%

23.Qwen 3.8 27B36.06%

24.Gemini 3.7 Flash34.62%

25.GLM-5.231.25%

26.Gemini 3.5 Flash30.77%

27.GPT-6 Luna30.29%

28.MiniMax M329.81%

29.GPT-6 Sol28.85%

30.Inkling28.37%

31.GLM-5.127.89%

32.Inkling Small25.48%

32.Qwen 3.7 Max25.48%

34.Gemini 3.6 Flash25%

35.DeepSeek-V4-Pro23.08%

36.Gemini 3.1 Pro Preview20.67%

37.Gemini 3 Flash Preview18.27%

38.Qwen 3.7 Plus16.35%

39.Kimi K2.515.87%

39.Kimi K2.615.87%

39.MiMo-V2.5-Pro15.87%

42.Grok 4.315.39%

42.Nemotron 3 Ultra15.39%

44.Gemini 3.5 Flash-Lite13.94%

44.Grok 4.20 Reasoning13.94%

46.GPT-5.4 mini12.5%

47.MiniMax M2.710.58%

48.MiMo-V2.59.14%

48.Mistral Medium 3.59.14%

50.GPT-5.4 nano6.25%

51.Gemini 3.1 Flash-Lite3.37%

52.Nemotron 3.5 Lightning2.89%

53.Ling 3.0 Flash0%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-legal-research:1:overall
Metric: all-pass rate
Scale: 0-100
Participants: 53
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 6749.531. Cost per task in USD: 23.610262. Source ID: vals-ai. Evidence capture ID: vals-legal_research-2026-09-23-c163b99e1a0e. Source payload hash: c163b99e1a0e2be1a20e1f44d2914a932e31049d0b5c6767a7803cb35e0b3a74. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

Health

MedCode

Overall

1.Claude Opus 563.57%

2.Gemini 3.1 Pro Preview59.06%

3.Claude Fable 556.07%

4.Gemini 3 Flash Preview55.92%

5.Gemini 3.5 Flash55.83%

6.Claude Opus 4.754.86%

7.Claude Fable 5.153.51%

8.Gemini 3.7 Flash53.39%

9.Claude Opus 4.853.22%

10.Gemini 3.6 Flash53.15%

11.Gemini 2.5 Pro50.59%

12.Claude Opus 5.549.8%

13.GPT-549.63%

14.Grok 4.749.55%

15.GPT-5.549.1%

16.Kimi K348.88%

17.GPT-6 Astra48.49%

18.Claude Opus 4.648.24%

19.Gemini 3.8 Flash48.14%

20.Gemini 3.1 Flash-Lite47.6%

21.Claude Sonnet 547.54%

22.o347.29%

23.GPT-6 Sol47.07%

24.MiniMax M346.29%

25.Claude Opus 4.545.17%

26.Grok 4.644.71%

27.GPT-6 Luna44.69%

28.GPT-5.6 Sol43.97%

29.Gemini 3.5 Flash-Lite43.49%

30.GPT-5.6 Terra43.41%

31.Grok 4.543.29%

32.GPT-5 mini43.05%

33.GLM-5.342.86%

34.GPT-5.6 Luna42.39%

35.GLM-5.141.6%

36.GPT-5.441.29%

37.Inkling41.19%

38.DeepSeek-V4.1-Flash41.17%

39.GPT-5.4 nano41.03%

40.GLM-5.240.77%

41.Qwen 3.8 Max 090240.67%

42.Claude Sonnet 4.540.57%

43.DeepSeek-V4-Pro40.46%

44.Kimi K2.640.14%

45.Kimi K2.539.32%

46.Qwen 3.7 Max38.75%

47.Nemotron 3 Ultra38.62%

48.Grok 4.338.07%

49.Inkling Small37.89%

50.MiniMax M2.734.44%

51.MiniMax M2.134.08%

52.o4-mini33.79%

53.Mistral Medium 3.533.75%

54.GLM-4.732.77%

55.MiMo-V2.5-Pro32.48%

56.Ling 3.0 Flash32.28%

57.Grok 4.20 Reasoning32.16%

58.MiMo-V2.531.9%

59.GPT-5 nano30.44%

60.Qwen 3.8 27B28.7%

61.Gemini 2.5 Flash-Lite27.12%

62.Llama 4 Scout23.31%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-medcode:1:overall
Metric: accuracy
Scale: 0-100
Participants: 62
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 246.876. Cost per task in USD: 0.658237. Source ID: vals-ai. Evidence capture ID: vals-medcode-2026-09-23-f51ab09f75f1. Source payload hash: f51ab09f75f1f2a43e4795332d87d10d469da5713ee9e294dfd8a018c22ed0b0. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

MedScribe

Overall

1.Claude Opus 5.591.43%

2.Claude Fable 5.191.29%

3.Claude Opus 590.99%

4.Grok 4.789.38%

5.GLM-5.3-Flash88.94%

6.GLM-5.388.81%

7.Claude Fable 588.52%

8.Kimi K387.96%

9.GPT-6 Astra87.91%

10.MiniMax M387.25%

11.Grok 4.586.88%

12.GPT-5.586.87%

13.Claude Opus 4.686.74%

14.Grok 4.686.53%

15.Claude Opus 4.885.76%

16.DeepSeek-V4.1-Flash85.5%

17.Inkling85.41%

18.GPT-5.6 Sol85.23%

19.Qwen 3.8 Max 090284.95%

20.Claude Sonnet 4.584.52%

21.Gemini 3.8 Flash84.5%

22.GPT-5.6 Luna84.39%

23.Inkling Small84.11%

24.Gemini 3.7 Flash83.94%

25.Qwen 3.8 27B83.85%

26.MiMo-V2.5-Pro83.73%

27.GPT-6 Luna83.71%

28.GPT-583.65%

29.GLM-5.283.53%

30.Claude Opus 4.583.25%

31.Claude Opus 4.782.95%

32.GPT-5.6 Terra82.87%

33.GPT-6 Sol82.03%

34.Ling 3.0 Flash80.9%

35.MiniMax M2.180.78%

36.GPT-5 mini80.58%

37.MiniMax M2.779.87%

38.Gemini 3.6 Flash79.66%

39.Qwen 3.7 Max79.4%

40.Kimi K2.678.15%

41.GPT-5.477.55%

42.GPT-5.4 nano77.09%

43.o376.65%

44.Gemini 3.5 Flash76.57%

45.Kimi K2.576.44%

46.Gemini 3.1 Pro Preview76.11%

47.Claude Sonnet 576.05%

48.DeepSeek-V4-Pro75.14%

49.Grok 4.374.4%

50.Gemini 2.5 Pro73.55%

51.GPT-5 nano72.87%

52.Gemini 2.5 Flash-Lite72.83%

53.GLM-5.172.27%

54.MiMo-V2.572.15%

55.Gemini 3.5 Flash-Lite70.89%

56.Gemini 3 Flash Preview69.92%

57.o4-mini69.14%

58.GLM-4.768.63%

59.Mistral Medium 3.567.73%

60.Gemini 3.1 Flash-Lite63.9%

61.Grok 4.20 Reasoning63.41%

62.Llama 4 Scout50.59%

63.Nemotron 3.5 Lightning4.27%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-medscribe:1:overall
Metric: accuracy
Scale: 0-100
Participants: 63
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 441.582. Cost per task in USD: 1.154156. Source ID: vals-ai. Evidence capture ID: vals-medscribe-2026-09-23-57321027150b. Source payload hash: 57321027150b04b875133d62749af33afaf3506f3c2f125e8b1b8707f46c4eaf. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

ProofBench

Overall

1.Claude Fable 5.1100%

1.Claude Opus 5.5100%

3.Claude Opus 599%

3.GPT-6 Astra99%

5.Claude Fable 595%

6.Kimi K387%

7.GPT-5.6 Sol83%

7.GPT-6 Sol83%

9.Claude Sonnet 577%

10.GPT-5.6 Terra74%

11.GPT-6 Luna64%

12.GPT-5.6 Luna60%

13.Gemini 3.7 Flash58%

13.Qwen 3.8 Max 090258%

15.DeepSeek-V4.1-Flash54%

16.Grok 4.651%

17.GLM-5.349%

18.Gemini 3.8 Flash48%

19.Gemini 3.5 Flash31%

19.Grok 4.531%

21.Gemini 3.1 Pro Preview26%

21.Grok 4.726%

23.MiMo-V2.5-Pro22%

24.GLM-5.3-Flash21%

25.MiniMax M318%

26.DeepSeek-V4-Pro16%

26.MiMo-V2.516%

26.Qwen 3.8 27B16%

29.Mistral Medium 3.59%

30.Inkling Small6%

31.Inkling0%

higher is betterAs of September 23, 2026Scale: 0% - 100%

Source: Vals AI

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: vals-proof-bench:1.1:overall
Metric: accuracy
Scale: 0-100
Participants: 31
Methodology and context: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 351.035. Cost per task in USD: 0.924063. Source ID: vals-ai. Evidence capture ID: vals-proof_bench-2026-09-23-323050c29bf5. Source payload hash: 323050c29bf5c9f0fca38f6a87aec1f91b031cf350e783c60a892de3e2671cb0. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source: Vals AI

Recorded measurement details

Every observation retains its source value and published test conditions.

BioMysteryBench: Overall (Show measurement, test conditions, and source)

Source value: 79.259
Score: 79.259
Metric: accuracy
Unit: %
Benchmark version: 1
Category: general
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-biomysterybench:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-22. Harness: Terminus 2. Latency in seconds: 739.252. Cost per task in USD: 2.972555. Source ID: vals-ai. Evidence capture ID: vals-biomysterybench-2026-09-23-1da8cd6684f0. Source payload hash: 1da8cd6684f0e235fa297f17a076add04c169bb4739ba048aa3300d9e5028eaa. Parser version: vals-astro-v2. Imported at: 2026-09-23
Comparison methodology: Harness: Terminus 2

Source Vals AI

Code Migration: Overall (Show measurement, test conditions, and source)

Source value: 66.646
Score: 66.646
Metric: accuracy
Unit: %
Benchmark version: 1
Category: coding
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-code-migration:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 11925.532. Cost per task in USD: 69.480849. Source ID: vals-ai. Evidence capture ID: vals-code-migration-2026-09-23-c7fade0e4383. Source payload hash: c7fade0e4383b27bca9fb2e591abb1b60950f35fca099c1ae51cad7c14fb7292. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

EMB: Overall (Show measurement, test conditions, and source)

Source value: 75.938
Score: 75.938
Metric: accuracy
Unit: %
Benchmark version: 1
Category: finance
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-emb:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 2106.434. Cost per task in USD: 9.330415. Source ID: vals-ai. Evidence capture ID: vals-emb-2026-09-23-0e761c761c4e. Source payload hash: 0e761c761c4e6aa3d255a9a37e58f22915cd69f36faede07660f9a663ff4a6dc. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

Finance Agent (v2): Overall (Show measurement, test conditions, and source)

Source value: 58.587
Score: 58.587
Metric: accuracy
Unit: %
Benchmark version: 2
Category: finance
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-finance-agent-v2:2:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 1987.81. Cost per task in USD: 9.025984. Source ID: vals-ai. Evidence capture ID: vals-fabv2-2026-09-23-8253d7a09218. Source payload hash: 8253d7a092188677e18f64165f4f338e5b99be804498c7d5476f877cd246914b. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

IOI: Overall (Show measurement, test conditions, and source)

Source value: 95.056
Score: 95.056
Metric: accuracy
Unit: %
Benchmark version: 2
Category: coding
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-ioi:2:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-22. Latency in seconds: 1035.191. Cost per task in USD: 4.382385. Source ID: vals-ai. Evidence capture ID: vals-ioi-2026-09-23-81487e138796. Source payload hash: 81487e1387966f49c2cce85653523e20447fd6ca8cee8e30372da5b934f8fa0c. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

Harvey's Legal Agent Benchmark: Overall · Task fully resolved (Show measurement, test conditions, and source)

Source value: 3.75
Score: 3.75
Metric: task resolution rate
Unit: %
Benchmark version: 1
Category: legal
Task: overall
Task label: Overall · Task fully resolved
Direction: Higher is better
Comparison cohort: vals-legal-agent-benchmark:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-22. Latency in seconds: 2939.571. Cost per task in USD: 17.569061. Source ID: vals-ai. Evidence capture ID: vals-hlab-2026-09-23-507154b1cf0d. Source payload hash: 507154b1cf0dcf67b4d36cb1325228e600b8f727addbed9e862cdc0b61b3af0c. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

Legal Research Bench: Overall · All-pass (Show measurement, test conditions, and source)

Source value: 50.481
Score: 50.481
Metric: all-pass rate
Unit: %
Benchmark version: 1
Category: legal
Task: overall
Task label: Overall · All-pass
Direction: Higher is better
Comparison cohort: vals-legal-research:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 6749.531. Cost per task in USD: 23.610262. Source ID: vals-ai. Evidence capture ID: vals-legal_research-2026-09-23-c163b99e1a0e. Source payload hash: c163b99e1a0e2be1a20e1f44d2914a932e31049d0b5c6767a7803cb35e0b3a74. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

MedCode: Overall (Show measurement, test conditions, and source)

Source value: 49.797
Score: 49.797
Metric: accuracy
Unit: %
Benchmark version: 1
Category: health
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-medcode:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 246.876. Cost per task in USD: 0.658237. Source ID: vals-ai. Evidence capture ID: vals-medcode-2026-09-23-f51ab09f75f1. Source payload hash: f51ab09f75f1f2a43e4795332d87d10d469da5713ee9e294dfd8a018c22ed0b0. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

MedScribe: Overall (Show measurement, test conditions, and source)

Source value: 91.43
Score: 91.43
Metric: accuracy
Unit: %
Benchmark version: 1
Category: health
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-medscribe:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 441.582. Cost per task in USD: 1.154156. Source ID: vals-ai. Evidence capture ID: vals-medscribe-2026-09-23-57321027150b. Source payload hash: 57321027150b04b875133d62749af33afaf3506f3c2f125e8b1b8707f46c4eaf. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

MysteryMechanism: Overall (Show measurement, test conditions, and source)

Source value: 49.55
Score: 49.55
Metric: accuracy
Unit: %
Benchmark version: 1
Category: general
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-mysterymechanism:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 1178.454. Cost per task in USD: 4.059711. Source ID: vals-ai. Evidence capture ID: vals-mysterymechanism-2026-09-23-63be0832dc5b. Source payload hash: 63be0832dc5bfc3769dae7055df0a6f6c1b46b0da25663f9022faff971e28703. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

ProgramBench: Overall (Show measurement, test conditions, and source)

Source value: 18.5
Score: 18.5
Metric: accuracy
Unit: %
Benchmark version: 1
Category: coding
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-programbench:1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-09-22. Latency in seconds: 8660.211. Cost per task in USD: 42.65866. Source ID: vals-ai. Evidence capture ID: vals-programbench-2026-09-23-083c733e7512. Source payload hash: 083c733e7512bc651ac31bb15b24fc2b0e6881e24a20748e847265e5556a012a. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

ProofBench: Overall (Show measurement, test conditions, and source)

Source value: 100
Score: 100
Metric: accuracy
Unit: %
Benchmark version: 1.1
Category: math
Task: overall
Task label: Overall
Direction: Higher is better
Comparison cohort: vals-proof-bench:1.1:overall
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-09-23
Methodology: Source model ID: anthropic/claude-opus-5-5. Source model name: Claude Opus 5.5. Evaluator: Vals AI. Dataset type: private. Leaderboard date: 2026-09-22. Latency in seconds: 351.035. Cost per task in USD: 0.924063. Source ID: vals-ai. Evidence capture ID: vals-proof_bench-2026-09-23-323050c29bf5. Source payload hash: 323050c29bf5c9f0fca38f6a87aec1f91b031cf350e783c60a892de3e2671cb0. Parser version: vals-astro-v2. Imported at: 2026-09-23

Source Vals AI

Page 1 of 3

Profile

Specifications and access

Published information about this model. Existing estimates are explicitly labeled.

Dated documentation and model-card observations. Provider limits, native context and extended context can differ. Configuration notes retain the source wording.

Specification	Value and source
Weights	ProprietarySource
Access	APISource
Context window	1,000,000 tokensSource
Knowledge cutoff	June 2026Source
Notes	Released September 22, 2026. Anthropic positions Opus 5.5 for long-running agentic coding and knowledge work, priced at $4 / $20 per million input and output tokens. 1M context, 128K output, a June 2026 knowledge cutoff, adaptive thinking that is always on, and medium effort by default. Official API ID: claude-opus-5-5.Source
API model ID	claude-opus-5-5Source
Source status	activeSource
Release	2026-09-22Source
Maximum output	128,000 tokensSource
Input	Text, imagesSource
Output	TextSource
Reasoning mode	adaptive (always on); default mediumSource
Knowledge cutoff	2026-06Source

Model generations

Predecessor and successor

Every relationship names its kind, its qualification and the dated source it rests on.

Evidence

Sources and data date

Every statement links to its underlying documentation or leaderboard.

Type	Evidence and data date
Price	Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. Source: Pricing · Model metadata (retrieved September 22, 2026)
Input	Documented modalities. Retrieved September 22, 2026. Source: Model documentation · Model metadata (retrieved September 22, 2026)
Output	Documented modalities. Retrieved September 22, 2026. Source: Model documentation · Model metadata (retrieved September 22, 2026)
Research status	Research date September 22, 2026. 2 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources	parameters-billions Original wording Anthropic does not publish parameter counts for this model.
Not found in the inspected sources	active-parameters-billions Original wording Anthropic does not publish active parameter counts for this model.
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Vals AI (retrieved September 23, 2026)
Additional source	Anthropic (retrieved September 22, 2026)

Claude Opus 5.5

Published benchmarks

General tasks

BioMysteryBench

MysteryMechanism

Coding

Code Migration

IOI

ProgramBench

Finance

EMB

Finance Agent (v2)

Legal

Harvey's Legal Agent Benchmark

Legal Research Bench

Health

MedCode

MedScribe

ProofBench

Recorded measurement details

Specifications and access

Published prices

Compare this model

Predecessor and successor

Sources and data date