Benchmark Almanac

Vals Index v1

Claude Opus 4.7 72.2% leads since April 2026
Kimi K3 74.7% leads since July 2026
Only in system cards

Vals AI's first index of agentic work in finance, law and coding, with each sector weighted by its share of U.S. GDP.

Versions 1.0 to 1.2 ran from October 2025 to 12 August 2026. Until May 2026 it combined CorpFin, Finance Agent, CaseLaw, SWE-bench Verified and Terminal-Bench 2.0. In May 2026 Vals dropped the law sector, added Vibe Code Bench, moved to Finance Agent v2 and to Terminal-Bench 2.1. The archived v1.2 is (2.0 × finance + 1.4 × coding) / 3.4.

Made by
Vals AI
Released
October 2025
Kind
Composite work indices
Scored
Graded by a model
Answers
Agent task
Tasks
Public, plus a private test set
Strongest model tested
GPT-5.5 #20 on the capability index
Last in a system card
July 2026
Last model added
July 2026
Scores refreshed
November 12th, 2026
35
Entries on its own leaderboard, 34 models we track
1
System card reports it, from 1 lab

Claude Opus 4.7 leads with 72.2% among models released by April 2026

Best score per entry 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

Kimi K3 leads with 74.7%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

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

35 entries. Scroll sideways for the rest.

  1. 72.2 Claude Opus 4.7
  2. 71.0 GPT-5.5
  3. 66.1 GPT-5.4
  4. 66.0 Claude Opus 4.6
  5. 65.5 Claude Sonnet 4.6
  6. 64.5 GPT-5.2
  7. 62.2 DeepSeek-V4-Pro
  8. 59.3 Kimi K2.6
  9. 58.1 Gemini 3.1 Pro
  10. 57.9 GLM-5.1
  11. 56.6 Muse Spark
  12. 55.7 GPT-5.4 Mini
  13. 55.1 GLM-5
  14. 54.6 Claude Opus 4.5
  15. 53.6 GPT-5.1
  16. 53.3 Qwen 3.6 Plus
  17. 53.2 Claude Sonnet 4.5
  18. 53.2 Gemini 3 Flash
  19. 52.5 Grok 4.3
  20. 52.1 Gemini 3 Pro
  21. 51.9 GPT-5.4 Nano
  22. 51.7 Kimi K2.5
  23. 50.8 Qwen 3.5 Plus (hosted 397B-A17B)
  24. 49.9 MiniMax-M2.7
  25. 49.0 GPT-5
  26. 48.7 Qwen3.6 27B
  27. 47.2 Claude Haiku 4.5
  28. 46.1 Grok 4.20
  29. 44.5 GPT-5 mini
  30. 42.6 Grok 4 Fast
  31. 42.5 Grok 4.1 Fast
  32. 40.8 Gemini 3.1 Flash-Lite
  33. 39.3 Gemini 2.5 Pro (snapshot unspecified)
  34. 29.1 Mistral Small 4
  35. 26.3 DeepSeek-V3.2
  1. 1 Claude Opus 4.7 72.2%
  2. 2 GPT-5.5 71%
  3. 3 GPT-5.4 xhigh 66.1%
  4. 4 Claude Opus 4.6 thinking 66%
  5. 5 Claude Sonnet 4.6 65.5%
  6. 6 GPT-5.2 64.5%
  7. 7 DeepSeek-V4-Pro 62.2%
  8. 8 Kimi K2.6 59.3%
  9. 9 Gemini 3.1 Pro 58.1%
  10. 10 GLM-5.1 57.9%
  11. 11 Muse Spark 56.6%
  12. 12 GPT-5.4 Mini 55.7%
  13. 13 GLM-5 thinking 55.1%
  14. 14 Claude Opus 4.5 thinking 54.6%
  15. 15 GPT-5.1 53.6%
  16. 16 Qwen 3.6 Plus 53.3%
  17. 17 Claude Sonnet 4.5 thinking 53.2%
  18. 18 Gemini 3 Flash 53.2%
  19. 19 Grok 4.3 52.5%
  20. 20 Gemini 3 Pro 52.1%
  21. 21 GPT-5.4 Nano 51.9%
  22. 22 Kimi K2.5 thinking 51.7%
  23. 23 Qwen 3.5 Plus (hosted 397B-A17B) thinking 50.8%
  24. 24 MiniMax-M2.7 49.9%
  25. 25 GPT-5 49%
  26. 26 Qwen3.6 27B 48.7%
  27. 27 Claude Haiku 4.5 thinking 47.2%
  28. 28 Grok 4.20 reasoning 46.1%
  29. 29 GPT-5 mini 44.5%
  30. 30 Grok 4 Fast reasoning 42.6%
  31. 31 Grok 4.1 Fast reasoning 42.5%
  32. 32 Gemini 3.1 Flash-Lite 40.8%
  33. 33 Gemini 2.5 Pro (snapshot unspecified) 39.3%
  34. 34 Mistral Small 4 29.1%
  35. 35 DeepSeek-V3.2 thinking 26.3%

Source: Official Vals Index v1 leaderboard

35 models. Scroll sideways for the rest.

  1. 74.7 Kimi K3
  2. 72.2 Claude Opus 4.7
  3. 71.0 GPT-5.5
  4. 66.1 GPT-5.4
  5. 66.0 Claude Opus 4.6
  6. 65.5 Claude Sonnet 4.6
  7. 64.5 GPT-5.2
  8. 62.2 DeepSeek-V4-Pro
  9. 59.3 Kimi K2.6
  10. 58.1 Gemini 3.1 Pro
  11. 57.9 GLM-5.1
  12. 56.6 Muse Spark
  13. 55.7 GPT-5.4 Mini
  14. 55.1 GLM-5
  15. 54.6 Claude Opus 4.5
  16. 53.6 GPT-5.1
  17. 53.3 Qwen 3.6 Plus
  18. 53.2 Claude Sonnet 4.5
  19. 53.2 Gemini 3 Flash
  20. 52.5 Grok 4.3
  21. 52.1 Gemini 3 Pro
  22. 51.9 GPT-5.4 Nano
  23. 51.7 Kimi K2.5
  24. 50.8 Qwen 3.5 Plus (hosted 397B-A17B)
  25. 49.9 MiniMax-M2.7
  26. 49.0 GPT-5
  27. 48.7 Qwen3.6 27B
  28. 47.2 Claude Haiku 4.5
  29. 46.1 Grok 4.20
  30. 44.5 GPT-5 mini
  31. 42.6 Grok 4 Fast
  32. 42.5 Grok 4.1 Fast
  33. 40.8 Gemini 3.1 Flash-Lite
  34. 29.1 Mistral Small 4
  35. 26.3 DeepSeek-V3.2
  1. 1 Kimi K3 74.7%
  2. 2 Claude Opus 4.7 72.2%
  3. 3 GPT-5.5 71%
  4. 4 GPT-5.4 xhigh 66.1%
  5. 5 Claude Opus 4.6 thinking 66%
  6. 6 Claude Sonnet 4.6 65.5%
  7. 7 GPT-5.2 64.5%
  8. 8 DeepSeek-V4-Pro 62.2%
  9. 9 Kimi K2.6 59.3%
  10. 10 Gemini 3.1 Pro 58.1%
  11. 11 GLM-5.1 57.9%
  12. 12 Muse Spark 56.6%
  13. 13 GPT-5.4 Mini 55.7%
  14. 14 GLM-5 thinking 55.1%
  15. 15 Claude Opus 4.5 thinking 54.6%
  16. 16 GPT-5.1 53.6%
  17. 17 Qwen 3.6 Plus 53.3%
  18. 18 Claude Sonnet 4.5 thinking 53.2%
  19. 19 Gemini 3 Flash 53.2%
  20. 20 Grok 4.3 52.5%
  21. 21 Gemini 3 Pro 52.1%
  22. 22 GPT-5.4 Nano 51.9%
  23. 23 Kimi K2.5 thinking 51.7%
  24. 24 Qwen 3.5 Plus (hosted 397B-A17B) thinking 50.8%
  25. 25 MiniMax-M2.7 49.9%
  26. 26 GPT-5 49%
  27. 27 Qwen3.6 27B 48.7%
  28. 28 Claude Haiku 4.5 thinking 47.2%
  29. 29 Grok 4.20 reasoning 46.1%
  30. 30 GPT-5 mini 44.5%
  31. 31 Grok 4 Fast reasoning 42.6%
  32. 32 Grok 4.1 Fast reasoning 42.5%
  33. 33 Gemini 3.1 Flash-Lite 40.8%
  34. 34 Mistral Small 4 29.1%
  35. 35 DeepSeek-V3.2 thinking 26.3%

Sources: Labs’ model cards and official Vals Index v1 leaderboard

#EntryScoreSettingDateSource
1 Claude Opus 4.7 72.2% Apr 2026 Official leaderboard
2 GPT-5.5 71% Apr 2026 Official leaderboard
3 GPT-5.4 66.1% xhigh Mar 2026 Official leaderboard
4 Claude Opus 4.6 66% thinking Feb 2026 Official leaderboard
5 Claude Sonnet 4.6 65.5% Feb 2026 Official leaderboard
6 GPT-5.2 64.5% Dec 2025 Official leaderboard
7 DeepSeek-V4-Pro 62.2% Apr 2026 Official leaderboard
8 Kimi K2.6 59.3% Apr 2026 Official leaderboard
9 Gemini 3.1 Pro 58.1% Feb 2026 Official leaderboard
10 GLM-5.1 57.9% Apr 2026 Official leaderboard
11 Muse Spark 56.6% Apr 2026 Official leaderboard
12 GPT-5.4 Mini 55.7% Mar 2026 Official leaderboard
13 GLM-5 55.1% thinking Feb 2026 Official leaderboard
14 Claude Opus 4.5 54.6% thinking Nov 2025 Official leaderboard
15 GPT-5.1 53.6% Nov 2025 Official leaderboard
16 Qwen 3.6 Plus 53.3% Mar 2026 Official leaderboard
17 Claude Sonnet 4.5 53.2% thinking Sep 2025 Official leaderboard
18 Gemini 3 Flash 53.2% Dec 2025 Official leaderboard
19 Grok 4.3 52.5% Apr 2026 Official leaderboard
20 Gemini 3 Pro 52.1% Nov 2025 Official leaderboard
21 GPT-5.4 Nano 51.9% Mar 2026 Official leaderboard
22 Kimi K2.5 51.7% thinking Jan 2026 Official leaderboard
23 Qwen 3.5 Plus (hosted 397B-A17B) 50.8% thinking Feb 2026 Official leaderboard
24 MiniMax-M2.7 49.9% Mar 2026 Official leaderboard
25 GPT-5 49% Aug 2025 Official leaderboard
26 Qwen3.6 27B 48.7% Apr 2026 Official leaderboard
27 Claude Haiku 4.5 47.2% thinking Oct 2025 Official leaderboard
28 Grok 4.20 46.1% reasoning Feb 2026 Official leaderboard
29 GPT-5 mini 44.5% Aug 2025 Official leaderboard
30 Grok 4 Fast 42.6% reasoning Sep 2025 Official leaderboard
31 Grok 4.1 Fast 42.5% reasoning Nov 2025 Official leaderboard
32 Gemini 3.1 Flash-Lite 40.8% Mar 2026 Official leaderboard
33 Gemini 2.5 Pro (snapshot unspecified) 39.3% Official leaderboard
34 Mistral Small 4 29.1% Mar 2026 Official leaderboard
35 DeepSeek-V3.2 26.3% thinking Dec 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Kimi K3 74.7% Jul 2026 Lab model card
2 Claude Opus 4.7 72.2% Apr 2026 Official leaderboard
3 GPT-5.5 71% Apr 2026 Official leaderboard
4 GPT-5.4 66.1% xhigh Mar 2026 Official leaderboard
5 Claude Opus 4.6 66% thinking Feb 2026 Official leaderboard
6 Claude Sonnet 4.6 65.5% Feb 2026 Official leaderboard
7 GPT-5.2 64.5% Dec 2025 Official leaderboard
8 DeepSeek-V4-Pro 62.2% Apr 2026 Official leaderboard
9 Kimi K2.6 59.3% Apr 2026 Official leaderboard
10 Gemini 3.1 Pro 58.1% Feb 2026 Official leaderboard
11 GLM-5.1 57.9% Apr 2026 Official leaderboard
12 Muse Spark 56.6% Apr 2026 Official leaderboard
13 GPT-5.4 Mini 55.7% Mar 2026 Official leaderboard
14 GLM-5 55.1% thinking Feb 2026 Official leaderboard
15 Claude Opus 4.5 54.6% thinking Nov 2025 Official leaderboard
16 GPT-5.1 53.6% Nov 2025 Official leaderboard
17 Qwen 3.6 Plus 53.3% Mar 2026 Official leaderboard
18 Claude Sonnet 4.5 53.2% thinking Sep 2025 Official leaderboard
19 Gemini 3 Flash 53.2% Dec 2025 Official leaderboard
20 Grok 4.3 52.5% Apr 2026 Official leaderboard
21 Gemini 3 Pro 52.1% Nov 2025 Official leaderboard
22 GPT-5.4 Nano 51.9% Mar 2026 Official leaderboard
23 Kimi K2.5 51.7% thinking Jan 2026 Official leaderboard
24 Qwen 3.5 Plus (hosted 397B-A17B) 50.8% thinking Feb 2026 Official leaderboard
25 MiniMax-M2.7 49.9% Mar 2026 Official leaderboard
26 GPT-5 49% Aug 2025 Official leaderboard
27 Qwen3.6 27B 48.7% Apr 2026 Official leaderboard
28 Claude Haiku 4.5 47.2% thinking Oct 2025 Official leaderboard
29 Grok 4.20 46.1% reasoning Feb 2026 Official leaderboard
30 GPT-5 mini 44.5% Aug 2025 Official leaderboard
31 Grok 4 Fast 42.6% reasoning Sep 2025 Official leaderboard
32 Grok 4.1 Fast 42.5% reasoning Nov 2025 Official leaderboard
33 Gemini 3.1 Flash-Lite 40.8% Mar 2026 Official leaderboard
34 Mistral Small 4 29.1% Mar 2026 Official leaderboard
35 DeepSeek-V3.2 26.3% thinking Dec 2025 Official leaderboard

Labs report scores for 1 model in their cards. Switch the scores above to rank them together.

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$10 Cost per test ($), log scale, lower is better Accuracy GPT-5: Accuracy 49%, cost per test $0.58 Claude Sonnet 4.5 (thinking): Accuracy 53.2%, cost per test $1.94 Claude Haiku 4.5 (thinking): Accuracy 47.2%, cost per test $0.5 GPT-5.1: Accuracy 53.6%, cost per test $0.87 Gemini 3 Pro: Accuracy 52.1%, cost per test $1.7 Grok 4.1 Fast (reasoning): Accuracy 42.5%, cost per test $0.07 Claude Opus 4.5 (thinking): Accuracy 54.6%, cost per test $7.44 DeepSeek-V3.2 (thinking): Accuracy 26.3%, cost per test $0.73 GPT-5.2: Accuracy 64.5%, cost per test $4.21 GLM-5 (thinking): Accuracy 55.1%, cost per test $8.33 Qwen 3.5 Plus (hosted 397B-A17B) (thinking): Accuracy 50.8%, cost per test $1.07 Grok 4.20 (reasoning): Accuracy 46.1%, cost per test $0.4 Gemini 3.1 Pro: Accuracy 58.1%, cost per test $1.26 Gemini 3.1 Flash-Lite: Accuracy 40.8%, cost per test $0.16 Mistral Small 4: Accuracy 29.1%, cost per test $1.39 GPT-5.4 Mini: Accuracy 55.7%, cost per test $0.66 MiniMax-M2.7: Accuracy 49.9%, cost per test $0.72 Qwen 3.6 Plus: Accuracy 53.3%, cost per test $1.34 GLM-5.1: Accuracy 57.9%, cost per test $0.8 Qwen3.6 27B: Accuracy 48.7%, cost per test $1.29 Grok 4.3: Accuracy 52.5%, cost per test $0.63 Gemini 2.5 Pro (snapshot unspecified): Accuracy 39.3%, cost per test $0.51. Not in our model list GPT-5 mini: Accuracy 44.5%, cost per test $0.12 Grok 4 Fast (reasoning): Accuracy 42.6%, cost per test $0.06 Gemini 3 Flash: Accuracy 53.2%, cost per test $0.36 Kimi K2.5 (thinking): Accuracy 51.7%, cost per test $0.29 Claude Opus 4.6 (thinking): Accuracy 66%, cost per test $2.5 Claude Sonnet 4.6: Accuracy 65.5%, cost per test $1.95 GPT-5.4 (xhigh): Accuracy 66.1%, cost per test $2.79 GPT-5.4 Nano: Accuracy 51.9%, cost per test $0.33 Claude Opus 4.7: Accuracy 72.2%, cost per test $5.51 Kimi K2.6: Accuracy 59.3%, cost per test $0.59 GPT-5.5: Accuracy 71%, cost per test $4.8 DeepSeek-V4-Pro: Accuracy 62.2%, cost per test $0.76 Claude Opus 4.7GPT-5.5GPT-5.4 (xhigh)Claude Opus 4.6 (thinking)Claude Sonnet 4.6DeepSeek-V4-ProKimi K2.6Gemini 3 FlashGPT-5.4 NanoKimi K2.5 (thinking)GPT-5 miniGrok 4 Fast (reasoning) 0% 20% 40% 60% 80% $0.01$0.1$1$10 Cost per test ($), log scale, lower is better Accuracy GPT-5: Accuracy 49%, cost per test $0.58 Claude Sonnet 4.5 (thinking): Accuracy 53.2%, cost per test $1.94 Claude Haiku 4.5 (thinking): Accuracy 47.2%, cost per test $0.5 GPT-5.1: Accuracy 53.6%, cost per test $0.87 Gemini 3 Pro: Accuracy 52.1%, cost per test $1.7 Grok 4.1 Fast (reasoning): Accuracy 42.5%, cost per test $0.07 Claude Opus 4.5 (thinking): Accuracy 54.6%, cost per test $7.44 DeepSeek-V3.2 (thinking): Accuracy 26.3%, cost per test $0.73 GPT-5.2: Accuracy 64.5%, cost per test $4.21 GLM-5 (thinking): Accuracy 55.1%, cost per test $8.33 Qwen 3.5 Plus (hosted 397B-A17B) (thinking): Accuracy 50.8%, cost per test $1.07 Grok 4.20 (reasoning): Accuracy 46.1%, cost per test $0.4 Gemini 3.1 Pro: Accuracy 58.1%, cost per test $1.26 Gemini 3.1 Flash-Lite: Accuracy 40.8%, cost per test $0.16 Mistral Small 4: Accuracy 29.1%, cost per test $1.39 GPT-5.4 Mini: Accuracy 55.7%, cost per test $0.66 MiniMax-M2.7: Accuracy 49.9%, cost per test $0.72 Qwen 3.6 Plus: Accuracy 53.3%, cost per test $1.34 GLM-5.1: Accuracy 57.9%, cost per test $0.8 Qwen3.6 27B: Accuracy 48.7%, cost per test $1.29 Grok 4.3: Accuracy 52.5%, cost per test $0.63 Gemini 2.5 Pro (snapshot unspecified): Accuracy 39.3%, cost per test $0.51. Not in our model list GPT-5 mini: Accuracy 44.5%, cost per test $0.12 Grok 4 Fast (reasoning): Accuracy 42.6%, cost per test $0.06 Gemini 3 Flash: Accuracy 53.2%, cost per test $0.36 Kimi K2.5 (thinking): Accuracy 51.7%, cost per test $0.29 Claude Opus 4.6 (thinking): Accuracy 66%, cost per test $2.5 Claude Sonnet 4.6: Accuracy 65.5%, cost per test $1.95 GPT-5.4 (xhigh): Accuracy 66.1%, cost per test $2.79 GPT-5.4 Nano: Accuracy 51.9%, cost per test $0.33 Claude Opus 4.7: Accuracy 72.2%, cost per test $5.51 Kimi K2.6: Accuracy 59.3%, cost per test $0.59 GPT-5.5: Accuracy 71%, cost per test $4.8 DeepSeek-V4-Pro: Accuracy 62.2%, cost per test $0.76 Claude Opus 4.7GPT-5.5GPT-5.4 (xhigh)Claude Opus 4.6 (thinking)Claude Sonnet 4.6DeepSeek-V4-ProKimi K2.6Gemini 3 FlashGPT-5.4 NanoKimi K2.5 (thinking)GPT-5 miniGrok 4 Fast (reasoning)
ModelSettingAccuracyCost per test
Claude Opus 4.7 72.2% $5.51
GPT-5.5 71% $4.8
GPT-5.4 xhigh 66.1% $2.79
Claude Opus 4.6 thinking 66% $2.5
Claude Sonnet 4.6 65.6% $1.95
GPT-5.2 64.5% $4.21
DeepSeek-V4-Pro 62.2% $0.76
Kimi K2.6 59.3% $0.59
Gemini 3.1 Pro 58.1% $1.26
GLM-5.1 57.9% $0.8
GPT-5.4 Mini 55.7% $0.66
GLM-5 thinking 55.1% $8.33
Claude Opus 4.5 thinking 54.6% $7.44
GPT-5.1 53.7% $0.87
Qwen 3.6 Plus 53.3% $1.34
Claude Sonnet 4.5 thinking 53.2% $1.94
Gemini 3 Flash 53.2% $0.36
Grok 4.3 52.5% $0.63
Gemini 3 Pro 52.1% $1.7
GPT-5.4 Nano 51.9% $0.33
Kimi K2.5 thinking 51.7% $0.29
Qwen 3.5 Plus (hosted 397B-A17B) thinking 50.8% $1.07
MiniMax-M2.7 49.9% $0.72
GPT-5 49% $0.58
Qwen3.6 27B 48.7% $1.29
Claude Haiku 4.5 thinking 47.2% $0.5
Grok 4.20 reasoning 46.1% $0.4
GPT-5 mini 44.5% $0.12
Grok 4 Fast reasoning 42.6% $0.06
Grok 4.1 Fast reasoning 42.5% $0.07
Gemini 3.1 Flash-Lite 40.8% $0.16
Gemini 2.5 Pro (snapshot unspecified) 39.3% $0.51
Mistral Small 4 29.1% $1.39
DeepSeek-V3.2 thinking 26.3% $0.73

The best score, 72.2%, costs $5.51 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.

OpenAI xAI Google DeepMind Moonshot AI Anthropic DeepSeek Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Opus 4.7 GPT-5.4, xhigh DeepSeek-V4-Pro Gemini 3 Flash GPT-5 mini Grok 4 Fast, reasoning 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Claude Opus 4.7 GPT-5.4, xhigh

Source: Official Vals Index v1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 4.7 72.2% $5.51
GPT-5.5 71% $4.80
GPT-5.4 xhigh 66.1% $2.79
Claude Opus 4.6 thinking 66% $2.50
Claude Sonnet 4.6 65.5% $1.95
GPT-5.2 64.5% $4.21
DeepSeek-V4-Pro 62.2% $0.76
Kimi K2.6 59.3% $0.59
Gemini 3.1 Pro 58.1% $1.26
GLM-5.1 57.9% $0.80
Muse Spark 56.6%
GPT-5.4 Mini 55.7% $0.66
GLM-5 thinking 55.1% $8.33
Claude Opus 4.5 thinking 54.6% $7.44
GPT-5.1 53.6% $0.87
Qwen 3.6 Plus 53.3% $1.34
Claude Sonnet 4.5 thinking 53.2% $1.94
Gemini 3 Flash 53.2% $0.36
Grok 4.3 52.5% $0.63
Gemini 3 Pro 52.1% $1.70
GPT-5.4 Nano 51.9% $0.33
Kimi K2.5 thinking 51.7% $0.29
Qwen 3.5 Plus (hosted 397B-A17B) thinking 50.8% $1.07
MiniMax-M2.7 49.9% $0.72
GPT-5 49% $0.58
Qwen3.6 27B 48.7% $1.29
Claude Haiku 4.5 thinking 47.2% $0.50
Grok 4.20 reasoning 46.1% $0.40
GPT-5 mini 44.5% $0.12
Grok 4 Fast reasoning 42.6% $0.06
Grok 4.1 Fast reasoning 42.5% $0.07
Gemini 3.1 Flash-Lite 40.8% $0.16
Mistral Small 4 29.1% $1.39
DeepSeek-V3.2 thinking 26.3% $0.73

7 models have held number one on the official leaderboard since its release in October 2025 8 models have held number one with claims included since its release in October 2025

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

Anthropic OpenAI
Own leaderboard Replaced by 2, Aug 2026
Claude Opus 4.7 72.2%

Source: Official Vals Index v1 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%) Kimi K3 Claude Opus 4.7 GPT-5.4 GPT-5.2 Claude Opus 4.5 GPT-5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Kimi K3 GPT-5.4

Sources: Official Vals Index v1 leaderboard and labs’ model cards

Labs report it only in system cards

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

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
Jul 27, 2026 Kimi K3 technical report 74.7 Vals Index, rank 2 of 39 (Table 5) Other

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

Replaced by Vals Index v2

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%