Benchmark Almanac

Vals Index v2.1

Gemini 4 Argon 68.9% leads since September 2026
In use

Version 2.1 of Vals AI's GDP-weighted index, which moved coding to Terminal-Bench 4.0 and added a tax sector.

Released 25 September 2026. Terminal-Bench 4.0 replaced 2.1, and Tax Agent Bench entered at weight 0.5, so the denominator rose from 14.8 to 15.3. Finance and legal were unchanged.

Made by
Vals AI
Released
September 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 in a launch post
October 2026
Last model added
October 2026
Scores refreshed
November 12th, 2026
49
Models scored on its own leaderboard
1
Launch post lists it, from 1 lab
1
System card reports it, from 1 lab

Gemini 4 Argon leads with 68.9%

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.

49 models. Scroll sideways for the rest.

  1. 68.9 Gemini 4 Argon
  2. 67.0 Claude Sonnet 5.5
  3. 67.0 Claude Opus 5.5
  4. 65.8 Claude Fable 5.1
  5. 64.6 Muse Wonder Full
  6. 63.7 Claude Opus 5
  7. 63.1 GPT-6 Astra
  8. 61.4 Claude Fable 5
  9. 61.1 GPT-6.1 Sol
  10. 61.1 Muse Wonder
  11. 58.2 Muse Spark 1.3
  12. 58.0 GPT-5.6 Sol
  13. 57.5 GPT-6 Sol
  14. 56.6 DeepSeek V4.1 Flash 1017
  15. 55.2 MiMo-V2.6 Pro
  16. 55.1 Claude Opus 4.8
  17. 55.0 Grok 4.7
  18. 54.9 Gemini 3.8 Flash
  19. 54.4 Claude Haiku 5.5
  20. 53.5 GLM-5.3
  21. 53.2 MiMo-V2.6 Flash
  22. 53.1 GPT-5.6 Terra
  23. 52.1 Grok 4.6
  24. 51.9 Claude Sonnet 5
  25. 51.7 GPT-5.6 Luna
  26. 51.3 DeepSeek V4.1 Flash
  27. 51.3 Gemini 3.7 Flash
  28. 51.2 GPT-6 Luna
  29. 50.9 Ember-1
  30. 50.3 Kimi K3
  31. 49.9 Tencent Hy4 preview
  32. 49.4 Step 5 Preview
  33. 49.3 Muse Spark 1.2
  34. 48.7 Mistral Large 4 Preview
  35. 48.3 Qwen 3.8 Max
  36. 48.0 DeepSeek V4 Flash 0731
  37. 47.6 DeepSeek V4 Pro 0813
  38. 46.6 Muse Spark Open
  39. 44.8 Gemini 3.5 Flash
  40. 44.7 Grok 4.5
  41. 44.6 GLM-5.3-Flash
  42. 38.6 DeepSeek-V4-Pro
  43. 36.5 MiniMax-M3
  44. 33.8 MiMo-V2.5-Pro
  45. 33.4 Gemini 3.1 Pro
  46. 33.2 GPT-5.4 Mini
  47. 28.7 Inkling
  48. 25.5 Inkling-Small
  49. 8.86 Mercury 2.5
  1. 1 Gemini 4 Argon 68.9%
  2. 2 Claude Sonnet 5.5 67%
  3. 3 Claude Opus 5.5 67%
  4. 4 Claude Fable 5.1 65.8%
  5. 5 Muse Wonder Full 64.6%
  6. 6 Claude Opus 5 63.7%
  7. 7 GPT-6 Astra 63.1%
  8. 8 Claude Fable 5 61.4%
  9. 9 GPT-6.1 Sol 61.1%
  10. 10 Muse Wonder 61.1%
  11. 11 Muse Spark 1.3 max 58.2%
  12. 12 GPT-5.6 Sol 58%
  13. 13 GPT-6 Sol 57.5%
  14. 14 DeepSeek V4.1 Flash 1017 56.6%
  15. 15 MiMo-V2.6 Pro 55.2%
  16. 16 Claude Opus 4.8 55.1%
  17. 17 Grok 4.7 55%
  18. 18 Gemini 3.8 Flash 54.9%
  19. 19 Claude Haiku 5.5 54.4%
  20. 20 GLM-5.3 53.5%
  21. 21 MiMo-V2.6 Flash 53.2%
  22. 22 GPT-5.6 Terra 53.1%
  23. 23 Grok 4.6 52.1%
  24. 24 Claude Sonnet 5 51.9%
  25. 25 GPT-5.6 Luna 51.7%
  26. 26 DeepSeek V4.1 Flash 51.3%
  27. 27 Gemini 3.7 Flash 51.3%
  28. 28 GPT-6 Luna 51.2%
  29. 29 Ember-1 50.9%
  30. 30 Kimi K3 50.3%
  31. 31 Tencent Hy4 preview 49.9%
  32. 32 Step 5 Preview 49.4%
  33. 33 Muse Spark 1.2 49.3%
  34. 34 Mistral Large 4 Preview 48.7%
  35. 35 Qwen 3.8 Max 48.3%
  36. 36 DeepSeek V4 Flash 0731 48%
  37. 37 DeepSeek V4 Pro 0813 47.6%
  38. 38 Muse Spark Open 46.6%
  39. 39 Gemini 3.5 Flash 44.8%
  40. 40 Grok 4.5 44.7%
  41. 41 GLM-5.3-Flash 44.6%
  42. 42 DeepSeek-V4-Pro 38.6%
  43. 43 MiniMax-M3 36.5%
  44. 44 MiMo-V2.5-Pro 33.8%
  45. 45 Gemini 3.1 Pro 33.4%
  46. 46 GPT-5.4 Mini 33.2%
  47. 47 Inkling 28.7%
  48. 48 Inkling-Small 25.5%
  49. 49 Mercury 2.5 8.86%

Source: Official Vals Index v2.1 leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Argon 68.9% Sep 2026 Official leaderboard
2 Claude Sonnet 5.5 67% Sep 2026 Official leaderboard
3 Claude Opus 5.5 67% Sep 2026 Official leaderboard
4 Claude Fable 5.1 65.8% Sep 2026 Official leaderboard
5 Muse Wonder Full 64.6% Oct 2026 Official leaderboard
6 Claude Opus 5 63.7% Jul 2026 Official leaderboard
7 GPT-6 Astra 63.1% Sep 2026 Official leaderboard
8 Claude Fable 5 61.4% Jun 2026 Official leaderboard
9 GPT-6.1 Sol 61.1% Sep 2026 Official leaderboard
10 Muse Wonder 61.1% Oct 2026 Official leaderboard
11 Muse Spark 1.3 58.2% max Sep 2026 Official leaderboard
12 GPT-5.6 Sol 58% Jul 2026 Official leaderboard
13 GPT-6 Sol 57.5% Sep 2026 Official leaderboard
14 DeepSeek V4.1 Flash 1017 56.6% Oct 2026 Official leaderboard
15 MiMo-V2.6 Pro 55.2% Sep 2026 Official leaderboard
16 Claude Opus 4.8 55.1% May 2026 Official leaderboard
17 Grok 4.7 55% Sep 2026 Official leaderboard
18 Gemini 3.8 Flash 54.9% Sep 2026 Official leaderboard
19 Claude Haiku 5.5 54.4% Oct 2026 Official leaderboard
20 GLM-5.3 53.5% Aug 2026 Official leaderboard
21 MiMo-V2.6 Flash 53.2% Sep 2026 Official leaderboard
22 GPT-5.6 Terra 53.1% Jul 2026 Official leaderboard
23 Grok 4.6 52.1% Aug 2026 Official leaderboard
24 Claude Sonnet 5 51.9% Jun 2026 Official leaderboard
25 GPT-5.6 Luna 51.7% Jul 2026 Official leaderboard
26 DeepSeek V4.1 Flash 51.3% Sep 2026 Official leaderboard
27 Gemini 3.7 Flash 51.3% Aug 2026 Official leaderboard
28 GPT-6 Luna 51.2% Sep 2026 Official leaderboard
29 Ember-1 50.9% Sep 2026 Official leaderboard
30 Kimi K3 50.3% Jul 2026 Official leaderboard
31 Tencent Hy4 preview 49.9% Aug 2026 Official leaderboard
32 Step 5 Preview 49.4% Sep 2026 Official leaderboard
33 Muse Spark 1.2 49.3% Aug 2026 Official leaderboard
34 Mistral Large 4 Preview 48.7% Oct 2026 Official leaderboard
35 Qwen 3.8 Max 48.3% Aug 2026 Official leaderboard
36 DeepSeek V4 Flash 0731 48% Jul 2026 Official leaderboard
37 DeepSeek V4 Pro 0813 47.6% Aug 2026 Official leaderboard
38 Muse Spark Open 46.6% Oct 2026 Official leaderboard
39 Gemini 3.5 Flash 44.8% May 2026 Official leaderboard
40 Grok 4.5 44.7% Jul 2026 Official leaderboard
41 GLM-5.3-Flash 44.6% Aug 2026 Official leaderboard
42 DeepSeek-V4-Pro 38.6% Apr 2026 Official leaderboard
43 MiniMax-M3 36.5% Jun 2026 Official leaderboard
44 MiMo-V2.5-Pro 33.8% Apr 2026 Official leaderboard
45 Gemini 3.1 Pro 33.4% Feb 2026 Official leaderboard
46 GPT-5.4 Mini 33.2% Mar 2026 Official leaderboard
47 Inkling 28.7% Jul 2026 Official leaderboard
48 Inkling-Small 25.5% Jul 2026 Official leaderboard
49 Mercury 2.5 8.86% Sep 2026 Official leaderboard

Labs report scores for 1 model in their cards. We plot every source in the charts below.

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% $0.1$1$10$100 Cost per test ($), log scale, lower is better Accuracy Gemini 3.1 Pro: Accuracy 33.4%, cost per test $2.18 GPT-5.4 Mini: Accuracy 33.2%, cost per test $1.45 DeepSeek-V4-Pro: Accuracy 38.6%, cost per test $1.31 Gemini 3.5 Flash: Accuracy 44.8%, cost per test $3.3 Claude Opus 4.8: Accuracy 55.1%, cost per test $13.1 MiniMax-M3: Accuracy 36.5%, cost per test $3.08 Claude Fable 5: Accuracy 61.4%, cost per test $29.6 Claude Sonnet 5: Accuracy 51.9%, cost per test $13.7 Grok 4.5: Accuracy 44.7%, cost per test $2.42 GPT-5.6 Luna: Accuracy 51.7%, cost per test $0.82 GPT-5.6 Sol: Accuracy 58%, cost per test $14.2 GPT-5.6 Terra: Accuracy 53.1%, cost per test $5.78 Inkling: Accuracy 28.7%, cost per test $1.46 Inkling-Small: Accuracy 25.5%, cost per test $0.28 Kimi K3: Accuracy 50.3%, cost per test $6.44 Claude Opus 5: Accuracy 63.7%, cost per test $19.3 DeepSeek V4 Flash 0731: Accuracy 48%, cost per test $0.85 Qwen 3.8 Max: Accuracy 48.3%, cost per test $4.96 Muse Spark 1.2: Accuracy 49.3%, cost per test $1.94 Grok 4.6: Accuracy 52.1%, cost per test $4.49 DeepSeek V4 Pro 0813: Accuracy 47.6%, cost per test $3.35 Gemini 3.7 Flash: Accuracy 51.3%, cost per test $4.67 GLM-5.3: Accuracy 53.5%, cost per test $7.25 GLM-5.3-Flash: Accuracy 44.6%, cost per test $1.03 Tencent Hy4 preview: Accuracy 49.9%, cost per test $1.44 Claude Fable 5.1: Accuracy 65.8%, cost per test $28.7 Gemini 3.8 Flash: Accuracy 54.9%, cost per test $5.71 Muse Spark 1.3 (max): Accuracy 58.2%, cost per test $3.79 Muse Spark 1.3: Accuracy 53.2%, cost per test $3.38 GPT-6 Astra: Accuracy 63.1%, cost per test $18.5 Mercury 2.5: Accuracy 8.86%, cost per test $0.29 DeepSeek V4.1 Flash: Accuracy 51.3%, cost per test $0.33 Step 5 Preview: Accuracy 49.4%, cost per test $2.63 Grok 4.7: Accuracy 55%, cost per test $12.1 Claude Opus 5.5: Accuracy 67%, cost per test $32.1 GPT-6 Luna: Accuracy 51.2%, cost per test $0.43 GPT-6 Sol: Accuracy 57.5%, cost per test $7.58 Ember-1: Accuracy 50.9%, cost per test $9.66 Claude Sonnet 5.5: Accuracy 67%, cost per test $21.3 Mistral Large 4 Preview: Accuracy 48.7%, cost per test $14.1 Claude Haiku 5.5: Accuracy 54.4%, cost per test $3 Muse Wonder: Accuracy 61.1%, cost per test $11.4 Muse Spark Open: Accuracy 46.6%, cost per test $4.08 MiMo-V2.5-Pro: Accuracy 33.8%, cost per test $0.16 MiMo-V2.6 Flash: Accuracy 53.2%, cost per test $0.2 MiMo-V2.6 Pro: Accuracy 55.2%, cost per test $0.41 GPT-6.1 Sol: Accuracy 61.1%, cost per test $3.24 Gemini 4 Argon: Accuracy 68.9%, cost per test $15.7 Muse Wonder Full: Accuracy 64.6%, cost per test $8.56 DeepSeek V4.1 Flash 1017: Accuracy 56.6%, cost per test $0.64 Gemini 4 ArgonMuse Wonder FullGPT-6.1 SolDeepSeek V4.1 Flash 1017MiMo-V2.6 ProMiMo-V2.6 FlashMiMo-V2.5-Pro 0% 20% 40% 60% 80% $0.1$1$10$100 Cost per test ($), log scale, lower is better Accuracy Gemini 3.1 Pro: Accuracy 33.4%, cost per test $2.18 GPT-5.4 Mini: Accuracy 33.2%, cost per test $1.45 DeepSeek-V4-Pro: Accuracy 38.6%, cost per test $1.31 Gemini 3.5 Flash: Accuracy 44.8%, cost per test $3.3 Claude Opus 4.8: Accuracy 55.1%, cost per test $13.1 MiniMax-M3: Accuracy 36.5%, cost per test $3.08 Claude Fable 5: Accuracy 61.4%, cost per test $29.6 Claude Sonnet 5: Accuracy 51.9%, cost per test $13.7 Grok 4.5: Accuracy 44.7%, cost per test $2.42 GPT-5.6 Luna: Accuracy 51.7%, cost per test $0.82 GPT-5.6 Sol: Accuracy 58%, cost per test $14.2 GPT-5.6 Terra: Accuracy 53.1%, cost per test $5.78 Inkling: Accuracy 28.7%, cost per test $1.46 Inkling-Small: Accuracy 25.5%, cost per test $0.28 Kimi K3: Accuracy 50.3%, cost per test $6.44 Claude Opus 5: Accuracy 63.7%, cost per test $19.3 DeepSeek V4 Flash 0731: Accuracy 48%, cost per test $0.85 Qwen 3.8 Max: Accuracy 48.3%, cost per test $4.96 Muse Spark 1.2: Accuracy 49.3%, cost per test $1.94 Grok 4.6: Accuracy 52.1%, cost per test $4.49 DeepSeek V4 Pro 0813: Accuracy 47.6%, cost per test $3.35 Gemini 3.7 Flash: Accuracy 51.3%, cost per test $4.67 GLM-5.3: Accuracy 53.5%, cost per test $7.25 GLM-5.3-Flash: Accuracy 44.6%, cost per test $1.03 Tencent Hy4 preview: Accuracy 49.9%, cost per test $1.44 Claude Fable 5.1: Accuracy 65.8%, cost per test $28.7 Gemini 3.8 Flash: Accuracy 54.9%, cost per test $5.71 Muse Spark 1.3 (max): Accuracy 58.2%, cost per test $3.79 Muse Spark 1.3: Accuracy 53.2%, cost per test $3.38 GPT-6 Astra: Accuracy 63.1%, cost per test $18.5 Mercury 2.5: Accuracy 8.86%, cost per test $0.29 DeepSeek V4.1 Flash: Accuracy 51.3%, cost per test $0.33 Step 5 Preview: Accuracy 49.4%, cost per test $2.63 Grok 4.7: Accuracy 55%, cost per test $12.1 Claude Opus 5.5: Accuracy 67%, cost per test $32.1 GPT-6 Luna: Accuracy 51.2%, cost per test $0.43 GPT-6 Sol: Accuracy 57.5%, cost per test $7.58 Ember-1: Accuracy 50.9%, cost per test $9.66 Claude Sonnet 5.5: Accuracy 67%, cost per test $21.3 Mistral Large 4 Preview: Accuracy 48.7%, cost per test $14.1 Claude Haiku 5.5: Accuracy 54.4%, cost per test $3 Muse Wonder: Accuracy 61.1%, cost per test $11.4 Muse Spark Open: Accuracy 46.6%, cost per test $4.08 MiMo-V2.5-Pro: Accuracy 33.8%, cost per test $0.16 MiMo-V2.6 Flash: Accuracy 53.2%, cost per test $0.2 MiMo-V2.6 Pro: Accuracy 55.2%, cost per test $0.41 GPT-6.1 Sol: Accuracy 61.1%, cost per test $3.24 Gemini 4 Argon: Accuracy 68.9%, cost per test $15.7 Muse Wonder Full: Accuracy 64.6%, cost per test $8.56 DeepSeek V4.1 Flash 1017: Accuracy 56.6%, cost per test $0.64 Gemini 4 ArgonMuse Wonder FullGPT-6.1 SolDeepSeek V4.1 Flash 1017MiMo-V2.6 ProMiMo-V2.6 FlashMiMo-V2.5-Pro
ModelSettingAccuracyCost per test
Gemini 4 Argon 68.9% $15.68
Claude Sonnet 5.5 67% $21.34
Claude Opus 5.5 67% $32.14
Claude Fable 5.1 65.8% $28.71
Muse Wonder Full 64.6% $8.56
Claude Opus 5 63.7% $19.31
GPT-6 Astra 63.1% $18.46
Claude Fable 5 61.4% $29.59
GPT-6.1 Sol 61.2% $3.24
Muse Wonder 61.1% $11.43
Muse Spark 1.3 max 58.2% $3.79
GPT-5.6 Sol 58% $14.24
GPT-6 Sol 57.5% $7.58
DeepSeek V4.1 Flash 1017 56.6% $0.64
MiMo-V2.6 Pro 55.2% $0.41
Claude Opus 4.8 55.1% $13.14
Grok 4.7 55% $12.11
Gemini 3.8 Flash 54.9% $5.71
Claude Haiku 5.5 54.4% $3
GLM-5.3 53.5% $7.25
MiMo-V2.6 Flash 53.2% $0.2
Muse Spark 1.3 53.2% $3.38
GPT-5.6 Terra 53.1% $5.78
Grok 4.6 52.1% $4.49
Claude Sonnet 5 51.9% $13.69
GPT-5.6 Luna 51.7% $0.82
DeepSeek V4.1 Flash 51.3% $0.33
Gemini 3.7 Flash 51.3% $4.67
GPT-6 Luna 51.2% $0.43
Ember-1 50.9% $9.66
Kimi K3 50.3% $6.44
Tencent Hy4 preview 49.9% $1.44
Step 5 Preview 49.4% $2.63
Muse Spark 1.2 49.3% $1.94
Mistral Large 4 Preview 48.7% $14.07
Qwen 3.8 Max 48.3% $4.96
DeepSeek V4 Flash 0731 48% $0.85
DeepSeek V4 Pro 0813 47.6% $3.35
Muse Spark Open 46.6% $4.08
Gemini 3.5 Flash 44.8% $3.3
Grok 4.5 44.7% $2.42
GLM-5.3-Flash 44.6% $1.03
DeepSeek-V4-Pro 38.6% $1.31
MiniMax-M3 36.5% $3.08
MiMo-V2.5-Pro 33.8% $0.16
Gemini 3.1 Pro 33.4% $2.18
GPT-5.4 Mini 33.2% $1.45
Inkling 28.7% $1.46
Inkling-Small 25.5% $0.28
Mercury 2.5 8.9% $0.29

The best score, 68.9%, costs $16 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 Google DeepMind Meta DeepSeek Cost frontier One model at each effort level
0% 20% 40% 60% 80% $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Argon Muse Wonder Full GPT-6.1 Sol DeepSeek V4.1 Flash 1017 MiMo-V2.6 Flash MiMo-V2.5-Pro 0% 20% 40% 60% 80% $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Argon Muse Wonder Full

Source: Official Vals Index v2.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Argon 68.9% $16
Claude Sonnet 5.5 67% $21
Claude Opus 5.5 67% $32
Claude Fable 5.1 65.8% $29
Muse Wonder Full 64.6% $8.56
Claude Opus 5 63.7% $19
GPT-6 Astra 63.1% $18
Claude Fable 5 61.4% $30
GPT-6.1 Sol 61.1% $3.24
Muse Wonder 61.1% $11
Muse Spark 1.3 max 58.2% $3.79
GPT-5.6 Sol 58% $14
GPT-6 Sol 57.5% $7.58
DeepSeek V4.1 Flash 1017 56.6% $0.64
MiMo-V2.6 Pro 55.2% $0.41
Claude Opus 4.8 55.1% $13
Grok 4.7 55% $12
Gemini 3.8 Flash 54.9% $5.71
Claude Haiku 5.5 54.4% $3.00
GLM-5.3 53.5% $7.25
MiMo-V2.6 Flash 53.2% $0.20
Muse Spark 1.3 53.2% $3.38
GPT-5.6 Terra 53.1% $5.78
Grok 4.6 52.1% $4.49
Claude Sonnet 5 51.9% $14
GPT-5.6 Luna 51.7% $0.82
DeepSeek V4.1 Flash 51.3% $0.33
Gemini 3.7 Flash 51.3% $4.67
GPT-6 Luna 51.2% $0.43
Ember-1 50.9% $9.66
Kimi K3 50.3% $6.44
Tencent Hy4 preview 49.9% $1.44
Step 5 Preview 49.4% $2.63
Muse Spark 1.2 49.3% $1.94
Mistral Large 4 Preview 48.7% $14
Qwen 3.8 Max 48.3% $4.96
DeepSeek V4 Flash 0731 48% $0.85
DeepSeek V4 Pro 0813 47.6% $3.35
Muse Spark Open 46.6% $4.08
Gemini 3.5 Flash 44.8% $3.30
Grok 4.5 44.7% $2.42
GLM-5.3-Flash 44.6% $1.03
DeepSeek-V4-Pro 38.6% $1.31
MiniMax-M3 36.5% $3.08
MiMo-V2.5-Pro 33.8% $0.16
Gemini 3.1 Pro 33.4% $2.18
GPT-5.4 Mini 33.2% $1.45
Inkling 28.7% $1.46
Inkling-Small 25.5% $0.28
Mercury 2.5 8.86% $0.29

4 models have held number one on the official leaderboard since its release in September 2026

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

Anthropic Google DeepMind
Own leaderboard 4 leaders
Gemini 4 Argon 68.9%

Source: Official Vals Index v2.1 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 Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Claude Opus 5 Claude Opus 4.8 Gemini 3.5 Flash DeepSeek-V4-Pro MiMo-V2.5-Pro Gemini 3.1 Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Claude Fable 5

Sources: Official Vals Index v2.1 leaderboard and labs’ model cards

One lab put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Google DeepMind
DateLaunch postScoreAs the post writes it
Oct 13, 2026 Gemini 4 Argon 68.9 Vals Index

One system card reports it

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Oct 13, 2026 Gemini 4 Argon model card 68.9 Vals Index (results table, p.5) Capabilities

One lab claim next to the leaderboard

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 1 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Gemini 4 Argon 68.9% Vals Index 68.9% Leaderboard Same

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 v2 Replaced by Vals Index v2.2

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%