Benchmark Almanac

Vals Index v2

Claude Opus 5.5 69.7% leads since September 2026
Not in model cards

Version 2 of Vals AI's GDP-weighted index, which brought back a legal sector and added financial modeling in Excel and code migration.

Released 13 August 2026. EMB replaced CorpFin, Code Migration joined coding, Legal Research Bench and HLAB formed the legal sector, and SWE-bench Verified left as saturated. Coding averages Terminal-Bench 2.1, Vibe Code Bench and Code Migration.

Made by
Vals AI
Released
August 2026
Kind
Composite work indices
Scored
Graded by a model
Answers
Agent task
Tasks
Public, plus a private test set
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
September 2026
Scores refreshed
November 12th, 2026
65
Models scored on its own leaderboard

Claude Opus 5.5 leads with 69.7% among models released by September 2026

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

65 models. Scroll sideways for the rest.

  1. 69.7 Claude Opus 5.5
  2. 69.2 Claude Sonnet 5.5
  3. 68.8 Claude Fable 5.1
  4. 67.2 Claude Opus 5
  5. 66.6 GPT-6 Astra
  6. 66.0 Claude Fable 5
  7. 64.5 Muse Spark 1.3
  8. 63.7 GPT-5.6 Sol
  9. 62.6 GPT-6 Sol
  10. 62.3 Gemini 3.8 Flash
  11. 60.9 Claude Opus 4.8
  12. 60.2 Grok 4.7
  13. 59.9 GPT-5.6 Luna
  14. 59.6 Claude Sonnet 5
  15. 59.6 GPT-5.6 Terra
  16. 59.6 MiMo-V2.6 Flash
  17. 59.5 MiMo-V2.6 Pro
  18. 59.3 Gemini 3.7 Flash
  19. 59.2 Grok 4.6
  20. 58.5 GPT-6 Luna
  21. 57.9 DeepSeek V4.1 Flash
  22. 57.8 Kimi K3
  23. 57.4 GPT-5.5
  24. 57.0 Muse Spark 1.2
  25. 57.0 GLM-5.3
  26. 56.1 Claude Opus 4.7
  27. 55.4 Tencent Hy4 preview
  28. 55.4 Gemini 3.6 Flash
  29. 54.8 Muse Spark 1.1
  30. 53.6 DeepSeek V4 Flash 0731
  31. 53.1 GLM-5.2
  32. 53.1 Gemini 3.5 Flash
  33. 52.4 DeepSeek V4 Pro 0813
  34. 51.8 Qwen 3.8 Max
  35. 51.5 Grok 4.5
  36. 50.6 Claude Sonnet 4.6
  37. 48.5 Qwen 3.8 27B
  38. 47.2 GLM-5.3-Flash
  39. 44.8 Qwen3.7-Max
  40. 43.5 Kimi K2.6
  41. 42.9 DeepSeek-V4-Pro
  42. 42.7 MiniMax-M3
  43. 41.9 Gemini 3.1 Pro
  44. 41.0 MiMo-V2.5-Pro
  45. 39.9 MiMo-V2.5
  46. 39.6 GPT-5.4 Mini
  47. 38.6 Qwen3.7-Plus
  48. 36.7 Gemini 3.5 Flash-Lite
  49. 34.1 Inkling
  50. 33.0 GPT-5.4 Nano
  51. 32.5 Ling-3.0-flash-Fin
  52. 32.0 Qwen 3.6 Plus
  53. 31.9 Inkling-Small
  54. 29.4 Gemini 3 Flash
  55. 27.4 Nemotron 3 Ultra
  56. 26.3 Kimi K2.5
  57. 24.6 MiniMax-M2.7
  58. 24.3 Grok 4.3
  59. 22.9 Claude Haiku 4.5
  60. 21.7 Ling-3.0-flash
  61. 17.9 Mistral Medium 3.5
  62. 17.6 Grok 4.20
  63. 15.5 Gemini 3.1 Flash-Lite
  64. 12.9 Mercury 2.5
  65. 11.5 Nemotron 3.5 Lightning 30B-A3B
  1. 1 Claude Opus 5.5 69.7%
  2. 2 Claude Sonnet 5.5 69.2%
  3. 3 Claude Fable 5.1 68.8%
  4. 4 Claude Opus 5 67.2%
  5. 5 GPT-6 Astra 66.6%
  6. 6 Claude Fable 5 66%
  7. 7 Muse Spark 1.3 max 64.5%
  8. 8 GPT-5.6 Sol 63.7%
  9. 9 GPT-6 Sol 62.6%
  10. 10 Gemini 3.8 Flash 62.3%
  11. 11 Claude Opus 4.8 60.9%
  12. 12 Grok 4.7 60.2%
  13. 13 GPT-5.6 Luna 59.9%
  14. 14 Claude Sonnet 5 59.6%
  15. 15 GPT-5.6 Terra 59.6%
  16. 16 MiMo-V2.6 Flash 59.6%
  17. 17 MiMo-V2.6 Pro 59.5%
  18. 18 Gemini 3.7 Flash 59.3%
  19. 19 Grok 4.6 59.2%
  20. 20 GPT-6 Luna 58.5%
  21. 21 DeepSeek V4.1 Flash 57.9%
  22. 22 Kimi K3 57.8%
  23. 23 GPT-5.5 57.4%
  24. 24 Muse Spark 1.2 57%
  25. 25 GLM-5.3 57%
  26. 26 Claude Opus 4.7 56.1%
  27. 27 Tencent Hy4 preview 55.4%
  28. 28 Gemini 3.6 Flash 55.4%
  29. 29 Muse Spark 1.1 54.8%
  30. 30 DeepSeek V4 Flash 0731 53.6%
  31. 31 GLM-5.2 53.1%
  32. 32 Gemini 3.5 Flash 53.1%
  33. 33 DeepSeek V4 Pro 0813 52.4%
  34. 34 Qwen 3.8 Max 51.8%
  35. 35 Grok 4.5 51.5%
  36. 36 Claude Sonnet 4.6 50.6%
  37. 37 Qwen 3.8 27B 48.5%
  38. 38 GLM-5.3-Flash 47.2%
  39. 39 Qwen3.7-Max 44.8%
  40. 40 Kimi K2.6 43.5%
  41. 41 DeepSeek-V4-Pro 42.9%
  42. 42 MiniMax-M3 42.7%
  43. 43 Gemini 3.1 Pro 41.9%
  44. 44 MiMo-V2.5-Pro 41%
  45. 45 MiMo-V2.5 39.9%
  46. 46 GPT-5.4 Mini 39.6%
  47. 47 Qwen3.7-Plus 38.6%
  48. 48 Gemini 3.5 Flash-Lite 36.7%
  49. 49 Inkling 34.1%
  50. 50 GPT-5.4 Nano 33%
  51. 51 Ling-3.0-flash-Fin 32.5%
  52. 52 Qwen 3.6 Plus 32%
  53. 53 Inkling-Small 31.9%
  54. 54 Gemini 3 Flash 29.4%
  55. 55 Nemotron 3 Ultra 27.4%
  56. 56 Kimi K2.5 thinking 26.3%
  57. 57 MiniMax-M2.7 24.6%
  58. 58 Grok 4.3 24.3%
  59. 59 Claude Haiku 4.5 thinking 22.9%
  60. 60 Ling-3.0-flash 21.7%
  61. 61 Mistral Medium 3.5 17.9%
  62. 62 Grok 4.20 reasoning 17.6%
  63. 63 Gemini 3.1 Flash-Lite 15.5%
  64. 64 Mercury 2.5 12.9%
  65. 65 Nemotron 3.5 Lightning 30B-A3B 11.5%

Source: Official Vals Index v2 leaderboard

#ModelScoreSettingDateSource
1 Claude Opus 5.5 69.7% Sep 2026 Official leaderboard
2 Claude Sonnet 5.5 69.2% Sep 2026 Official leaderboard
3 Claude Fable 5.1 68.8% Sep 2026 Official leaderboard
4 Claude Opus 5 67.2% Jul 2026 Official leaderboard
5 GPT-6 Astra 66.6% Sep 2026 Official leaderboard
6 Claude Fable 5 66% Jun 2026 Official leaderboard
7 Muse Spark 1.3 64.5% max Sep 2026 Official leaderboard
8 GPT-5.6 Sol 63.7% Jul 2026 Official leaderboard
9 GPT-6 Sol 62.6% Sep 2026 Official leaderboard
10 Gemini 3.8 Flash 62.3% Sep 2026 Official leaderboard
11 Claude Opus 4.8 60.9% May 2026 Official leaderboard
12 Grok 4.7 60.2% Sep 2026 Official leaderboard
13 GPT-5.6 Luna 59.9% Jul 2026 Official leaderboard
14 Claude Sonnet 5 59.6% Jun 2026 Official leaderboard
15 GPT-5.6 Terra 59.6% Jul 2026 Official leaderboard
16 MiMo-V2.6 Flash 59.6% Sep 2026 Official leaderboard
17 MiMo-V2.6 Pro 59.5% Sep 2026 Official leaderboard
18 Gemini 3.7 Flash 59.3% Aug 2026 Official leaderboard
19 Grok 4.6 59.2% Aug 2026 Official leaderboard
20 GPT-6 Luna 58.5% Sep 2026 Official leaderboard
21 DeepSeek V4.1 Flash 57.9% Sep 2026 Official leaderboard
22 Kimi K3 57.8% Jul 2026 Official leaderboard
23 GPT-5.5 57.4% Apr 2026 Official leaderboard
24 Muse Spark 1.2 57% Aug 2026 Official leaderboard
25 GLM-5.3 57% Aug 2026 Official leaderboard
26 Claude Opus 4.7 56.1% Apr 2026 Official leaderboard
27 Tencent Hy4 preview 55.4% Aug 2026 Official leaderboard
28 Gemini 3.6 Flash 55.4% Jul 2026 Official leaderboard
29 Muse Spark 1.1 54.8% Jul 2026 Official leaderboard
30 DeepSeek V4 Flash 0731 53.6% Jul 2026 Official leaderboard
31 GLM-5.2 53.1% Jun 2026 Official leaderboard
32 Gemini 3.5 Flash 53.1% May 2026 Official leaderboard
33 DeepSeek V4 Pro 0813 52.4% Aug 2026 Official leaderboard
34 Qwen 3.8 Max 51.8% Aug 2026 Official leaderboard
35 Grok 4.5 51.5% Jul 2026 Official leaderboard
36 Claude Sonnet 4.6 50.6% Feb 2026 Official leaderboard
37 Qwen 3.8 27B 48.5% Aug 2026 Official leaderboard
38 GLM-5.3-Flash 47.2% Aug 2026 Official leaderboard
39 Qwen3.7-Max 44.8% May 2026 Official leaderboard
40 Kimi K2.6 43.5% Apr 2026 Official leaderboard
41 DeepSeek-V4-Pro 42.9% Apr 2026 Official leaderboard
42 MiniMax-M3 42.7% Jun 2026 Official leaderboard
43 Gemini 3.1 Pro 41.9% Feb 2026 Official leaderboard
44 MiMo-V2.5-Pro 41% Apr 2026 Official leaderboard
45 MiMo-V2.5 39.9% Apr 2026 Official leaderboard
46 GPT-5.4 Mini 39.6% Mar 2026 Official leaderboard
47 Qwen3.7-Plus 38.6% Jun 2026 Official leaderboard
48 Gemini 3.5 Flash-Lite 36.7% Jul 2026 Official leaderboard
49 Inkling 34.1% Jul 2026 Official leaderboard
50 GPT-5.4 Nano 33% Mar 2026 Official leaderboard
51 Ling-3.0-flash-Fin 32.5% Sep 2026 Official leaderboard
52 Qwen 3.6 Plus 32% Mar 2026 Official leaderboard
53 Inkling-Small 31.9% Jul 2026 Official leaderboard
54 Gemini 3 Flash 29.4% Dec 2025 Official leaderboard
55 Nemotron 3 Ultra 27.4% Jun 2026 Official leaderboard
56 Kimi K2.5 26.3% thinking Jan 2026 Official leaderboard
57 MiniMax-M2.7 24.6% Mar 2026 Official leaderboard
58 Grok 4.3 24.3% Apr 2026 Official leaderboard
59 Claude Haiku 4.5 22.9% thinking Oct 2025 Official leaderboard
60 Ling-3.0-flash 21.7% Aug 2026 Official leaderboard
61 Mistral Medium 3.5 17.9% Apr 2026 Official leaderboard
62 Grok 4.20 17.6% reasoning Feb 2026 Official leaderboard
63 Gemini 3.1 Flash-Lite 15.5% Mar 2026 Official leaderboard
64 Mercury 2.5 12.9% Sep 2026 Official leaderboard
65 Nemotron 3.5 Lightning 30B-A3B 11.5% Aug 2026 Official leaderboard

Accuracy versus cost

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 80% $0.01$0.1$1$10$100 Cost per test ($), log scale, lower is better Accuracy Claude Haiku 4.5 (thinking): Accuracy 22.9%, cost per test $0.7 Gemini 3 Flash: Accuracy 29.4%, cost per test $0.41 Kimi K2.5 (thinking): Accuracy 26.3%, cost per test $0.51 Claude Sonnet 4.6: Accuracy 50.6%, cost per test $8.4 Grok 4.20 (reasoning): Accuracy 17.6%, cost per test $0.57 Gemini 3.1 Pro: Accuracy 41.9%, cost per test $1.94 Gemini 3.1 Flash-Lite: Accuracy 15.5%, cost per test $0.14 GPT-5.4 Mini: Accuracy 39.6%, cost per test $1.38 GPT-5.4 Nano: Accuracy 33%, cost per test $0.4 MiniMax-M2.7: Accuracy 24.6%, cost per test $0.88 Qwen 3.6 Plus: Accuracy 32%, cost per test $2.62 Claude Opus 4.7: Accuracy 56.1%, cost per test $12.8 Kimi K2.6: Accuracy 43.5%, cost per test $1.85 GPT-5.5: Accuracy 57.4%, cost per test $6.14 DeepSeek-V4-Pro: Accuracy 42.9%, cost per test $0.94 Mistral Medium 3.5: Accuracy 17.9%, cost per test $8.01 Grok 4.3: Accuracy 24.3%, cost per test $0.68 Gemini 3.5 Flash: Accuracy 53.1%, cost per test $2.91 Qwen3.7-Max: Accuracy 44.8%, cost per test $2.85 Claude Opus 4.8: Accuracy 60.9%, cost per test $12.7 MiniMax-M3: Accuracy 42.7%, cost per test $2.5 Qwen3.7-Plus: Accuracy 38.6%, cost per test $0.44 Claude Fable 5: Accuracy 66%, cost per test $28.7 GLM-5.2: Accuracy 53.1%, cost per test $4.29 Claude Sonnet 5: Accuracy 59.6%, cost per test $11.7 Grok 4.5: Accuracy 51.5%, cost per test $1.96 GPT-5.6 Sol: Accuracy 63.7%, cost per test $14.2 GPT-5.6 Terra: Accuracy 59.6%, cost per test $5.18 Muse Spark 1.1: Accuracy 54.8%, cost per test $1.25 Inkling: Accuracy 34.1%, cost per test $1.35 Inkling-Small: Accuracy 31.9%, cost per test $0.28 Kimi K3: Accuracy 57.8%, cost per test $6.47 Gemini 3.5 Flash-Lite: Accuracy 36.7%, cost per test $0.54 Gemini 3.6 Flash: Accuracy 55.4%, cost per test $3.03 DeepSeek V4 Flash 0731: Accuracy 53.6%, cost per test $0.84 Qwen 3.8 Max: Accuracy 51.8%, cost per test $4 Ling-3.0-flash: Accuracy 21.7%, cost per test $0.1 Muse Spark 1.2: Accuracy 57%, cost per test $1.66 Nemotron 3.5 Lightning 30B-A3B: Accuracy 11.5%, cost per test $0.09 Grok 4.6: Accuracy 59.2%, cost per test $4.34 DeepSeek V4 Pro 0813: Accuracy 52.4%, cost per test $3.28 Gemini 3.7 Flash: Accuracy 59.3%, cost per test $4.17 GLM-5.3: Accuracy 57%, cost per test $6.73 Qwen 3.8 27B: Accuracy 48.5%, cost per test $8.48 GLM-5.3-Flash: Accuracy 47.2%, cost per test $0.99 Tencent Hy4 preview: Accuracy 55.4%, cost per test $1.28 Claude Fable 5.1: Accuracy 68.8%, cost per test $28.9 Gemini 3.8 Flash: Accuracy 62.3%, cost per test $5.39 GPT-6 Astra: Accuracy 66.6%, cost per test $19.1 Ling-3.0-flash-Fin: Accuracy 32.5%, cost per test $0.11 Mercury 2.5: Accuracy 12.9%, cost per test $0.3 DeepSeek V4.1 Flash: Accuracy 57.9%, cost per test $0.3 Grok 4.7: Accuracy 60.2%, cost per test $11.2 MiMo-V2.6 Pro: Accuracy 59.5%, cost per test $0.39 GPT-6 Luna: Accuracy 58.5%, cost per test $0.42 GPT-6 Sol: Accuracy 62.6%, cost per test $7.56 MiMo-V2.5: Accuracy 39.9%, cost per test $0.06 MiMo-V2.5-Pro: Accuracy 41%, cost per test $0.15 GPT-5.6 Luna: Accuracy 59.9%, cost per test $0.78 Claude Opus 5: Accuracy 67.2%, cost per test $18.8 Muse Spark 1.3 (max): Accuracy 64.5%, cost per test $3.4 Muse Spark 1.3: Accuracy 60.3%, cost per test $2.9 MiMo-V2.6 Flash: Accuracy 59.6%, cost per test $0.2 Claude Opus 5.5: Accuracy 69.7%, cost per test $32.8 Claude Sonnet 5.5: Accuracy 69.2%, cost per test $20.8 Claude Opus 5.5Claude Sonnet 5.5Claude Opus 5Muse Spark 1.3 (max)Muse Spark 1.3GPT-5.6 LunaMiMo-V2.6 FlashMiMo-V2.5-ProMiMo-V2.5 0% 20% 40% 60% 80% $0.01$0.1$1$10$100 Cost per test ($), log scale, lower is better Accuracy Claude Haiku 4.5 (thinking): Accuracy 22.9%, cost per test $0.7 Gemini 3 Flash: Accuracy 29.4%, cost per test $0.41 Kimi K2.5 (thinking): Accuracy 26.3%, cost per test $0.51 Claude Sonnet 4.6: Accuracy 50.6%, cost per test $8.4 Grok 4.20 (reasoning): Accuracy 17.6%, cost per test $0.57 Gemini 3.1 Pro: Accuracy 41.9%, cost per test $1.94 Gemini 3.1 Flash-Lite: Accuracy 15.5%, cost per test $0.14 GPT-5.4 Mini: Accuracy 39.6%, cost per test $1.38 GPT-5.4 Nano: Accuracy 33%, cost per test $0.4 MiniMax-M2.7: Accuracy 24.6%, cost per test $0.88 Qwen 3.6 Plus: Accuracy 32%, cost per test $2.62 Claude Opus 4.7: Accuracy 56.1%, cost per test $12.8 Kimi K2.6: Accuracy 43.5%, cost per test $1.85 GPT-5.5: Accuracy 57.4%, cost per test $6.14 DeepSeek-V4-Pro: Accuracy 42.9%, cost per test $0.94 Mistral Medium 3.5: Accuracy 17.9%, cost per test $8.01 Grok 4.3: Accuracy 24.3%, cost per test $0.68 Gemini 3.5 Flash: Accuracy 53.1%, cost per test $2.91 Qwen3.7-Max: Accuracy 44.8%, cost per test $2.85 Claude Opus 4.8: Accuracy 60.9%, cost per test $12.7 MiniMax-M3: Accuracy 42.7%, cost per test $2.5 Qwen3.7-Plus: Accuracy 38.6%, cost per test $0.44 Claude Fable 5: Accuracy 66%, cost per test $28.7 GLM-5.2: Accuracy 53.1%, cost per test $4.29 Claude Sonnet 5: Accuracy 59.6%, cost per test $11.7 Grok 4.5: Accuracy 51.5%, cost per test $1.96 GPT-5.6 Sol: Accuracy 63.7%, cost per test $14.2 GPT-5.6 Terra: Accuracy 59.6%, cost per test $5.18 Muse Spark 1.1: Accuracy 54.8%, cost per test $1.25 Inkling: Accuracy 34.1%, cost per test $1.35 Inkling-Small: Accuracy 31.9%, cost per test $0.28 Kimi K3: Accuracy 57.8%, cost per test $6.47 Gemini 3.5 Flash-Lite: Accuracy 36.7%, cost per test $0.54 Gemini 3.6 Flash: Accuracy 55.4%, cost per test $3.03 DeepSeek V4 Flash 0731: Accuracy 53.6%, cost per test $0.84 Qwen 3.8 Max: Accuracy 51.8%, cost per test $4 Ling-3.0-flash: Accuracy 21.7%, cost per test $0.1 Muse Spark 1.2: Accuracy 57%, cost per test $1.66 Nemotron 3.5 Lightning 30B-A3B: Accuracy 11.5%, cost per test $0.09 Grok 4.6: Accuracy 59.2%, cost per test $4.34 DeepSeek V4 Pro 0813: Accuracy 52.4%, cost per test $3.28 Gemini 3.7 Flash: Accuracy 59.3%, cost per test $4.17 GLM-5.3: Accuracy 57%, cost per test $6.73 Qwen 3.8 27B: Accuracy 48.5%, cost per test $8.48 GLM-5.3-Flash: Accuracy 47.2%, cost per test $0.99 Tencent Hy4 preview: Accuracy 55.4%, cost per test $1.28 Claude Fable 5.1: Accuracy 68.8%, cost per test $28.9 Gemini 3.8 Flash: Accuracy 62.3%, cost per test $5.39 GPT-6 Astra: Accuracy 66.6%, cost per test $19.1 Ling-3.0-flash-Fin: Accuracy 32.5%, cost per test $0.11 Mercury 2.5: Accuracy 12.9%, cost per test $0.3 DeepSeek V4.1 Flash: Accuracy 57.9%, cost per test $0.3 Grok 4.7: Accuracy 60.2%, cost per test $11.2 MiMo-V2.6 Pro: Accuracy 59.5%, cost per test $0.39 GPT-6 Luna: Accuracy 58.5%, cost per test $0.42 GPT-6 Sol: Accuracy 62.6%, cost per test $7.56 MiMo-V2.5: Accuracy 39.9%, cost per test $0.06 MiMo-V2.5-Pro: Accuracy 41%, cost per test $0.15 GPT-5.6 Luna: Accuracy 59.9%, cost per test $0.78 Claude Opus 5: Accuracy 67.2%, cost per test $18.8 Muse Spark 1.3 (max): Accuracy 64.5%, cost per test $3.4 Muse Spark 1.3: Accuracy 60.3%, cost per test $2.9 MiMo-V2.6 Flash: Accuracy 59.6%, cost per test $0.2 Claude Opus 5.5: Accuracy 69.7%, cost per test $32.8 Claude Sonnet 5.5: Accuracy 69.2%, cost per test $20.8 Claude Opus 5.5Claude Sonnet 5.5Claude Opus 5Muse Spark 1.3 (max)Muse Spark 1.3GPT-5.6 LunaMiMo-V2.6 FlashMiMo-V2.5-ProMiMo-V2.5
ModelSettingAccuracyCost per test
Claude Opus 5.5 69.7% $32.77
Claude Sonnet 5.5 69.2% $20.8
Claude Fable 5.1 68.8% $28.92
Claude Opus 5 67.2% $18.81
GPT-6 Astra 66.6% $19.09
Claude Fable 5 66% $28.73
Muse Spark 1.3 max 64.5% $3.4
GPT-5.6 Sol 63.7% $14.21
GPT-6 Sol 62.6% $7.56
Gemini 3.8 Flash 62.3% $5.39
Claude Opus 4.8 60.9% $12.67
Muse Spark 1.3 60.3% $2.9
Grok 4.7 60.2% $11.16
GPT-5.6 Luna 59.9% $0.78
Claude Sonnet 5 59.6% $11.74
GPT-5.6 Terra 59.6% $5.18
MiMo-V2.6 Flash 59.6% $0.2
MiMo-V2.6 Pro 59.5% $0.39
Gemini 3.7 Flash 59.3% $4.17
Grok 4.6 59.2% $4.34
GPT-6 Luna 58.5% $0.42
DeepSeek V4.1 Flash 57.9% $0.3
Kimi K3 57.8% $6.47
GPT-5.5 57.4% $6.14
Muse Spark 1.2 57.1% $1.66
GLM-5.3 57% $6.73
Claude Opus 4.7 56.1% $12.76
Tencent Hy4 preview 55.4% $1.28
Gemini 3.6 Flash 55.4% $3.03
Muse Spark 1.1 54.8% $1.25
DeepSeek V4 Flash 0731 53.6% $0.84
GLM-5.2 53.1% $4.29
Gemini 3.5 Flash 53.1% $2.91
DeepSeek V4 Pro 0813 52.4% $3.28
Qwen 3.8 Max 51.8% $4
Grok 4.5 51.5% $1.96
Claude Sonnet 4.6 50.6% $8.4
Qwen 3.8 27B 48.5% $8.48
GLM-5.3-Flash 47.2% $0.99
Qwen3.7-Max 44.8% $2.85
Kimi K2.6 43.5% $1.85
DeepSeek-V4-Pro 42.9% $0.94
MiniMax-M3 42.7% $2.5
Gemini 3.1 Pro 41.9% $1.94
MiMo-V2.5-Pro 41% $0.15
MiMo-V2.5 39.9% $0.06
GPT-5.4 Mini 39.6% $1.38
Qwen3.7-Plus 38.7% $0.44
Gemini 3.5 Flash-Lite 36.7% $0.54
Inkling 34.1% $1.35
GPT-5.4 Nano 33% $0.4
Ling-3.0-flash-Fin 32.6% $0.11
Qwen 3.6 Plus 32% $2.62
Inkling-Small 31.9% $0.28
Gemini 3 Flash 29.4% $0.41
Kimi K2.5 thinking 26.3% $0.51
MiniMax-M2.7 24.6% $0.88
Grok 4.3 24.3% $0.68
Claude Haiku 4.5 thinking 22.9% $0.7
Ling-3.0-flash 21.7% $0.1
Mistral Medium 3.5 18% $8.01
Grok 4.20 reasoning 17.6% $0.57
Gemini 3.1 Flash-Lite 15.5% $0.14
Mercury 2.5 13% $0.3
Nemotron 3.5 Lightning 30B-A3B 11.5% $0.09

The best score, 69.7%, costs $33 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Xiaomi Corp OpenAI Anthropic Meta Cost frontier One model at each effort level
0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5.5 Muse Spark 1.3, max Muse Spark 1.3 GPT-5.6 Luna MiMo-V2.6 Flash MiMo-V2.5-Pro MiMo-V2.5 0% 20% 40% 60% 80% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5.5 Muse Spark 1.3, max

Source: Official Vals Index v2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 5.5 69.7% $33
Claude Sonnet 5.5 69.2% $21
Claude Fable 5.1 68.8% $29
Claude Opus 5 67.2% $19
GPT-6 Astra 66.6% $19
Claude Fable 5 66% $29
Muse Spark 1.3 max 64.5% $3.40
GPT-5.6 Sol 63.7% $14
GPT-6 Sol 62.6% $7.56
Gemini 3.8 Flash 62.3% $5.39
Claude Opus 4.8 60.9% $13
Muse Spark 1.3 60.3% $2.90
Grok 4.7 60.2% $11
GPT-5.6 Luna 59.9% $0.78
Claude Sonnet 5 59.6% $12
GPT-5.6 Terra 59.6% $5.18
MiMo-V2.6 Flash 59.6% $0.20
MiMo-V2.6 Pro 59.5% $0.39
Gemini 3.7 Flash 59.3% $4.17
Grok 4.6 59.2% $4.34
GPT-6 Luna 58.5% $0.42
DeepSeek V4.1 Flash 57.9% $0.30
Kimi K3 57.8% $6.47
GPT-5.5 57.4% $6.14
Muse Spark 1.2 57% $1.66
GLM-5.3 57% $6.73
Claude Opus 4.7 56.1% $13
Tencent Hy4 preview 55.4% $1.28
Gemini 3.6 Flash 55.4% $3.03
Muse Spark 1.1 54.8% $1.25
DeepSeek V4 Flash 0731 53.6% $0.84
GLM-5.2 53.1% $4.29
Gemini 3.5 Flash 53.1% $2.91
DeepSeek V4 Pro 0813 52.4% $3.28
Qwen 3.8 Max 51.8% $4.00
Grok 4.5 51.5% $1.96
Claude Sonnet 4.6 50.6% $8.40
Qwen 3.8 27B 48.5% $8.48
GLM-5.3-Flash 47.2% $0.99
Qwen3.7-Max 44.8% $2.85
Kimi K2.6 43.5% $1.85
DeepSeek-V4-Pro 42.9% $0.94
MiniMax-M3 42.7% $2.50
Gemini 3.1 Pro 41.9% $1.94
MiMo-V2.5-Pro 41% $0.15
MiMo-V2.5 39.9% $0.06
GPT-5.4 Mini 39.6% $1.38
Qwen3.7-Plus 38.6% $0.44
Gemini 3.5 Flash-Lite 36.7% $0.54
Inkling 34.1% $1.35
GPT-5.4 Nano 33% $0.40
Ling-3.0-flash-Fin 32.5% $0.11
Qwen 3.6 Plus 32% $2.62
Inkling-Small 31.9% $0.28
Gemini 3 Flash 29.4% $0.41
Nemotron 3 Ultra 27.4%
Kimi K2.5 thinking 26.3% $0.51
MiniMax-M2.7 24.6% $0.88
Grok 4.3 24.3% $0.68
Claude Haiku 4.5 thinking 22.9% $0.70
Ling-3.0-flash 21.7% $0.10
Mistral Medium 3.5 17.9% $8.01
Grok 4.20 reasoning 17.6% $0.57
Gemini 3.1 Flash-Lite 15.5% $0.14
Mercury 2.5 12.9% $0.30
Nemotron 3.5 Lightning 30B-A3B 11.5% $0.09

3 models have held number one on the official leaderboard since its release in August 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Own leaderboard Replaced by 2.1, Sep 2026
Claude Opus 5.5 69.7%

Source: Official Vals Index v2 leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Claude Fable 5 Claude Opus 4.8 Claude Opus 4.7 Claude Sonnet 4.6 Gemini 3 Flash Claude Haiku 4.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Claude Opus 4.8

Source: Official Vals Index v2 leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic OpenAI Google DeepMind
Vals Index v2.2 16 leaders
Claude Fable 5.2 71.3%
Vals Index v1 Replaced by 2, Aug 2026
Claude Opus 4.7 72.2%
Vals Index v2 Replaced by 2.1, Sep 2026
Claude Opus 5.5 69.7%
Vals Index v2.1 4 leaders
Gemini 4 Argon 68.9%
Vals Index v2.2 2 leaders
Claude Fable 5.2 71.3%

Source: Official leaderboards

Replaces Vals Index v1 Replaced by Vals Index v2.1

More in Economic work

All 97 in Economic work
  • JobBench In use 130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics. GLM-5.3 61.4%
  • GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate. Seed2.1 Pro 87.9%
  • GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
  • AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
  • CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Claude Fable 5 75.9%
  • OfficeQA In use Grounded reasoning over a large corpus of U.S. Treasury Bulletin documents: locate tables and compute precise numeric answers. Claude Opus 4.7 86.3%