Benchmark Almanac

Vals Legal Index

Muse Wonder Full 77.9% leads since October 2026
Muse Wonder Full 77.9% leads since October 2026
In use

Agents are scored on Legal Research Bench, HLAB-Vals-1.1 and Tax Agent Bench, mapped to practice areas weighted by share of US law-office revenue.

Practice areas without a covering benchmark (7% of revenue) are not scored. The board had 33 models on October 29th, 2026.

Made by
Vals AI
Released
October 2026
Kind
Composite work indices
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 13th, 2026
38
Models scored on its own leaderboard
1
Launch post lists it, from 1 lab

Muse Wonder Full leads with 77.9%

Best score per model from the benchmark's own leaderboard, read on November 13th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

Muse Wonder Full leads with 77.9%, 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.

38 models. Scroll sideways for the rest.

  1. 77.9 Muse Wonder Full
  2. 76.8 Gemini 4 Argon
  3. 75.9 Gemini 4 Carbon
  4. 75.7 Muse Ember
  5. 75.3 Muse Wonder
  6. 74.6 GLM-5.3
  7. 73.9 Qwen3.8 Max (0902)
  8. 73.9 Inkling 2 Small
  9. 73.6 Gemini 4 Hydrogen
  10. 72.7 Grok 4.7
  11. 72.4 Claude Opus 5.5
  12. 72.1 MiMo-V2.6 Pro
  13. 71.9 DeepSeek V4.1 Flash 1017
  14. 71.7 Kimi K3
  15. 71.7 Tencent Hy4 preview
  16. 71.0 GLM-5.3-Flash
  17. 70.5 Tencent Hy4
  18. 70.3 Claude Haiku 5.5
  19. 70.2 Step 5 Preview
  20. 69.8 Gemini 3.8 Flash
  21. 69.2 Claude Fable 5.2
  22. 69.0 DeepSeek V4 Pro 0813
  23. 69.0 MiMo-V2.6 Flash
  24. 68.7 GPT-6.1 Sol
  25. 68.5 Muse Spark Open
  26. 68.4 GPT-6 Astra
  27. 68.0 Claude Sonnet 5.5
  28. 67.8 Mistral Large 4
  29. 67.1 MiniMax-M3.1-Flash
  30. 67.0 Kimi K3.1 Pro
  31. 65.6 Mistral Large 4 Preview
  32. 63.0 Mercury 3
  33. 62.2 GPT-6 Luna
  34. 61.8 Beam 1.0
  35. 58.8 GPT-6.1 Terra
  36. 57.8 Inkling-Small
  37. 55.4 Nemotron 3.5 Super VL
  38. 54.2 Inkling
  1. 1 Muse Wonder Full 77.9%
  2. 2 Gemini 4 Argon 76.8%
  3. 3 Gemini 4 Carbon 75.9%
  4. 4 Muse Ember 75.7%
  5. 5 Muse Wonder 75.3%
  6. 6 GLM-5.3 74.6%
  7. 7 Qwen3.8 Max (0902) 73.9%
  8. 8 Inkling 2 Small 73.9%
  9. 9 Gemini 4 Hydrogen 73.6%
  10. 10 Grok 4.7 72.7%
  11. 11 Claude Opus 5.5 72.4%
  12. 12 MiMo-V2.6 Pro 72.1%
  13. 13 DeepSeek V4.1 Flash 1017 71.9%
  14. 14 Kimi K3 71.7%
  15. 15 Tencent Hy4 preview 71.7%
  16. 16 GLM-5.3-Flash 71%
  17. 17 Tencent Hy4 70.5%
  18. 18 Claude Haiku 5.5 70.3%
  19. 19 Step 5 Preview 70.2%
  20. 20 Gemini 3.8 Flash 69.8%
  21. 21 Claude Fable 5.2 69.2%
  22. 22 DeepSeek V4 Pro 0813 69%
  23. 23 MiMo-V2.6 Flash 69%
  24. 24 GPT-6.1 Sol 68.7%
  25. 25 Muse Spark Open 68.5%
  26. 26 GPT-6 Astra 68.4%
  27. 27 Claude Sonnet 5.5 68%
  28. 28 Mistral Large 4 67.8%
  29. 29 MiniMax-M3.1-Flash 67.1%
  30. 30 Kimi K3.1 Pro 67%
  31. 31 Mistral Large 4 Preview 65.6%
  32. 32 Mercury 3 63%
  33. 33 GPT-6 Luna 62.2%
  34. 34 Beam 1.0 61.8%
  35. 35 GPT-6.1 Terra 58.8%
  36. 36 Inkling-Small 57.8%
  37. 37 Nemotron 3.5 Super VL 55.4%
  38. 38 Inkling 54.2%

Source: Official Vals Legal Index leaderboard

38 models. Scroll sideways for the rest.

  1. 77.9 Muse Wonder Full
  2. 76.8 Gemini 4 Argon
  3. 75.9 Gemini 4 Carbon
  4. 75.7 Muse Ember
  5. 75.3 Muse Wonder
  6. 74.6 GLM-5.3
  7. 73.9 Qwen3.8 Max (0902)
  8. 73.9 Inkling 2 Small
  9. 73.6 Gemini 4 Hydrogen
  10. 72.7 Grok 4.7
  11. 72.4 Claude Opus 5.5
  12. 72.1 MiMo-V2.6 Pro
  13. 71.9 DeepSeek V4.1 Flash 1017
  14. 71.7 Kimi K3
  15. 71.7 Tencent Hy4 preview
  16. 71.0 GLM-5.3-Flash
  17. 70.5 Tencent Hy4
  18. 70.3 Claude Haiku 5.5
  19. 70.2 Step 5 Preview
  20. 69.8 Gemini 3.8 Flash
  21. 69.2 Claude Fable 5.2
  22. 69.0 DeepSeek V4 Pro 0813
  23. 69.0 MiMo-V2.6 Flash
  24. 68.7 GPT-6.1 Sol
  25. 68.5 Muse Spark Open
  26. 68.4 GPT-6 Astra
  27. 68.0 Claude Sonnet 5.5
  28. 67.8 Mistral Large 4
  29. 67.1 MiniMax-M3.1-Flash
  30. 67.0 Kimi K3.1 Pro
  31. 65.6 Mistral Large 4 Preview
  32. 63.0 Mercury 3
  33. 62.2 GPT-6 Luna
  34. 61.8 Beam 1.0
  35. 58.8 GPT-6.1 Terra
  36. 57.8 Inkling-Small
  37. 55.4 Nemotron 3.5 Super VL
  38. 54.2 Inkling
  1. 1 Muse Wonder Full 77.9%
  2. 2 Gemini 4 Argon 76.8%
  3. 3 Gemini 4 Carbon 75.9%
  4. 4 Muse Ember 75.7%
  5. 5 Muse Wonder 75.3%
  6. 6 GLM-5.3 74.6%
  7. 7 Qwen3.8 Max (0902) 73.9%
  8. 8 Inkling 2 Small 73.9%
  9. 9 Gemini 4 Hydrogen 73.6%
  10. 10 Grok 4.7 72.7%
  11. 11 Claude Opus 5.5 72.4%
  12. 12 MiMo-V2.6 Pro 72.1%
  13. 13 DeepSeek V4.1 Flash 1017 71.9%
  14. 14 Kimi K3 71.7%
  15. 15 Tencent Hy4 preview 71.7%
  16. 16 GLM-5.3-Flash 71%
  17. 17 Tencent Hy4 70.5%
  18. 18 Claude Haiku 5.5 70.3%
  19. 19 Step 5 Preview 70.2%
  20. 20 Gemini 3.8 Flash 69.8%
  21. 21 Claude Fable 5.2 69.2%
  22. 22 DeepSeek V4 Pro 0813 69%
  23. 23 MiMo-V2.6 Flash 69%
  24. 24 GPT-6.1 Sol 68.7%
  25. 25 Muse Spark Open 68.5%
  26. 26 GPT-6 Astra 68.4%
  27. 27 Claude Sonnet 5.5 68%
  28. 28 Mistral Large 4 67.8%
  29. 29 MiniMax-M3.1-Flash 67.1%
  30. 30 Kimi K3.1 Pro 67%
  31. 31 Mistral Large 4 Preview 65.6%
  32. 32 Mercury 3 63%
  33. 33 GPT-6 Luna 62.2%
  34. 34 Beam 1.0 61.8%
  35. 35 GPT-6.1 Terra 58.8%
  36. 36 Inkling-Small 57.8%
  37. 37 Nemotron 3.5 Super VL 55.4%
  38. 38 Inkling 54.2%

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

#ModelScoreSettingDateSource
1 Muse Wonder Full 77.9% Oct 2026 Official leaderboard
2 Gemini 4 Argon 76.8% Sep 2026 Official leaderboard
3 Gemini 4 Carbon 75.9% Oct 2026 Official leaderboard
4 Muse Ember 75.7% Nov 2026 Official leaderboard
5 Muse Wonder 75.3% Oct 2026 Official leaderboard
6 GLM-5.3 74.6% Aug 2026 Official leaderboard
7 Qwen3.8 Max (0902) 73.9% Sep 2026 Official leaderboard
8 Inkling 2 Small 73.9% Nov 2026 Official leaderboard
9 Gemini 4 Hydrogen 73.6% Nov 2026 Official leaderboard
10 Grok 4.7 72.7% Sep 2026 Official leaderboard
11 Claude Opus 5.5 72.4% Sep 2026 Official leaderboard
12 MiMo-V2.6 Pro 72.1% Sep 2026 Official leaderboard
13 DeepSeek V4.1 Flash 1017 71.9% Oct 2026 Official leaderboard
14 Kimi K3 71.7% Jul 2026 Official leaderboard
15 Tencent Hy4 preview 71.7% Aug 2026 Official leaderboard
16 GLM-5.3-Flash 71% Aug 2026 Official leaderboard
17 Tencent Hy4 70.5% Nov 2026 Official leaderboard
18 Claude Haiku 5.5 70.3% Oct 2026 Official leaderboard
19 Step 5 Preview 70.2% Sep 2026 Official leaderboard
20 Gemini 3.8 Flash 69.8% Sep 2026 Official leaderboard
21 Claude Fable 5.2 69.2% Nov 2026 Official leaderboard
22 DeepSeek V4 Pro 0813 69% Aug 2026 Official leaderboard
23 MiMo-V2.6 Flash 69% Sep 2026 Official leaderboard
24 GPT-6.1 Sol 68.7% Sep 2026 Official leaderboard
25 Muse Spark Open 68.5% Oct 2026 Official leaderboard
26 GPT-6 Astra 68.4% Sep 2026 Official leaderboard
27 Claude Sonnet 5.5 68% Sep 2026 Official leaderboard
28 Mistral Large 4 67.8% Oct 2026 Official leaderboard
29 MiniMax-M3.1-Flash 67.1% Oct 2026 Official leaderboard
30 Kimi K3.1 Pro 67% Oct 2026 Official leaderboard
31 Mistral Large 4 Preview 65.6% Oct 2026 Official leaderboard
32 Mercury 3 63% Oct 2026 Official leaderboard
33 GPT-6 Luna 62.2% Sep 2026 Official leaderboard
34 Beam 1.0 61.8% Oct 2026 Official leaderboard
35 GPT-6.1 Terra 58.8% Oct 2026 Official leaderboard
36 Inkling-Small 57.8% Jul 2026 Official leaderboard
37 Nemotron 3.5 Super VL 55.4% Oct 2026 Official leaderboard
38 Inkling 54.2% Jul 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Muse Wonder Full 77.9% Oct 2026 Official leaderboard
2 Gemini 4 Argon 76.8% Sep 2026 Official leaderboard
3 Gemini 4 Carbon 75.9% Oct 2026 Official leaderboard
4 Muse Ember 75.7% Nov 2026 Official leaderboard
5 Muse Wonder 75.3% Oct 2026 Official leaderboard
6 GLM-5.3 74.6% Aug 2026 Official leaderboard
7 Qwen3.8 Max (0902) 73.9% Sep 2026 Official leaderboard
8 Inkling 2 Small 73.9% Nov 2026 Lab model card
9 Gemini 4 Hydrogen 73.6% Nov 2026 Official leaderboard
10 Grok 4.7 72.7% Sep 2026 Official leaderboard
11 Claude Opus 5.5 72.4% Sep 2026 Official leaderboard
12 MiMo-V2.6 Pro 72.1% Sep 2026 Official leaderboard
13 DeepSeek V4.1 Flash 1017 71.9% Oct 2026 Official leaderboard
14 Kimi K3 71.7% Jul 2026 Official leaderboard
15 Tencent Hy4 preview 71.7% Aug 2026 Official leaderboard
16 GLM-5.3-Flash 71% Aug 2026 Official leaderboard
17 Tencent Hy4 70.5% Nov 2026 Official leaderboard
18 Claude Haiku 5.5 70.3% Oct 2026 Official leaderboard
19 Step 5 Preview 70.2% Sep 2026 Official leaderboard
20 Gemini 3.8 Flash 69.8% Sep 2026 Official leaderboard
21 Claude Fable 5.2 69.2% Nov 2026 Official leaderboard
22 DeepSeek V4 Pro 0813 69% Aug 2026 Official leaderboard
23 MiMo-V2.6 Flash 69% Sep 2026 Official leaderboard
24 GPT-6.1 Sol 68.7% Sep 2026 Official leaderboard
25 Muse Spark Open 68.5% Oct 2026 Official leaderboard
26 GPT-6 Astra 68.4% Sep 2026 Official leaderboard
27 Claude Sonnet 5.5 68% Sep 2026 Official leaderboard
28 Mistral Large 4 67.8% Oct 2026 Official leaderboard
29 MiniMax-M3.1-Flash 67.1% Oct 2026 Official leaderboard
30 Kimi K3.1 Pro 67% Oct 2026 Official leaderboard
31 Mistral Large 4 Preview 65.6% Oct 2026 Official leaderboard
32 Mercury 3 63% Oct 2026 Official leaderboard
33 GPT-6 Luna 62.2% Sep 2026 Official leaderboard
34 Beam 1.0 61.8% Oct 2026 Official leaderboard
35 GPT-6.1 Terra 58.8% Oct 2026 Official leaderboard
36 Inkling-Small 57.8% Jul 2026 Official leaderboard
37 Nemotron 3.5 Super VL 55.4% Oct 2026 Official leaderboard
38 Inkling 54.2% Jul 2026 Official leaderboard

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

Accuracy against latency

The Vals scatter chart with its x axis switched from Cost to Latency: mean time per test against accuracy, on a log scale. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

60% 80% 100% 101001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy Inkling: Accuracy 54.2%, latency (mean per test) 737 s Inkling-Small: Accuracy 57.8%, latency (mean per test) 538 s Kimi K3: Accuracy 71.7%, latency (mean per test) 1,911 s DeepSeek V4 Pro 0813: Accuracy 69%, latency (mean per test) 1,593 s GLM-5.3: Accuracy 74.6%, latency (mean per test) 2,691 s GLM-5.3-Flash: Accuracy 71%, latency (mean per test) 1,703 s Tencent Hy4 preview: Accuracy 71.7%, latency (mean per test) 2,764 s Qwen3.8 Max (0902): Accuracy 73.9%, latency (mean per test) 5,277 s Gemini 3.8 Flash: Accuracy 69.8%, latency (mean per test) 555 s GPT-6 Astra: Accuracy 68.4%, latency (mean per test) 1,448 s Step 5 Preview: Accuracy 70.2%, latency (mean per test) 1,483 s Grok 4.7: Accuracy 72.7%, latency (mean per test) 1,473 s MiMo-V2.6 Flash: Accuracy 69%, latency (mean per test) 1,100 s MiMo-V2.6 Pro: Accuracy 72.1%, latency (mean per test) 1,742 s Claude Opus 5.5: Accuracy 72.4%, latency (mean per test) 4,757 s GPT-6 Luna: Accuracy 62.2%, latency (mean per test) 1,950 s Claude Sonnet 5.5: Accuracy 68%, latency (mean per test) 4,516 s GPT-6.1 Sol: Accuracy 68.7%, latency (mean per test) 1,900 s Mistral Large 4: Accuracy 67.8%, latency (mean per test) 1,139 s Mistral Large 4 Preview: Accuracy 65.6%, latency (mean per test) 4,164 s Claude Haiku 5.5: Accuracy 70.3%, latency (mean per test) 3,525 s Muse Wonder: Accuracy 75.3%, latency (mean per test) 775 s Muse Spark Open: Accuracy 68.5%, latency (mean per test) 5,607 s Nemotron 3.5 Super VL: Accuracy 55.4%, latency (mean per test) 109 s DeepSeek V4.1 Flash 1017: Accuracy 71.9%, latency (mean per test) 1,567 s Mercury 3: Accuracy 63%, latency (mean per test) 534 s MiniMax-M3.1-Flash: Accuracy 67.1%, latency (mean per test) 2,124 s Kimi K3.1 Pro: Accuracy 67%, latency (mean per test) 4,710 s GPT-6.1 Terra: Accuracy 58.8%, latency (mean per test) 2,592 s Gemini 4 Carbon: Accuracy 75.9%, latency (mean per test) 2,307 s Tencent Hy4: Accuracy 70.5%, latency (mean per test) 3,477 s Inkling 2 Small: Accuracy 73.9%, latency (mean per test) 813 s Claude Fable 5.2: Accuracy 69.2%, latency (mean per test) 890 s Gemini 4 Hydrogen: Accuracy 73.6%, latency (mean per test) 1,267 s Gemini 4 Argon: Accuracy 76.8%, latency (mean per test) 1,541 s Beam 1.0: Accuracy 61.8%, latency (mean per test) 89.8 s Muse Wonder Full: Accuracy 77.9%, latency (mean per test) 1,796 s Muse Ember: Accuracy 75.7%, latency (mean per test) 217 s Muse Wonder FullGemini 4 ArgonMuse EmberBeam 1.0 60% 80% 100% 101001,00010K Latency (mean per test) (s), log scale, lower is better Accuracy Inkling: Accuracy 54.2%, latency (mean per test) 737 s Inkling-Small: Accuracy 57.8%, latency (mean per test) 538 s Kimi K3: Accuracy 71.7%, latency (mean per test) 1,911 s DeepSeek V4 Pro 0813: Accuracy 69%, latency (mean per test) 1,593 s GLM-5.3: Accuracy 74.6%, latency (mean per test) 2,691 s GLM-5.3-Flash: Accuracy 71%, latency (mean per test) 1,703 s Tencent Hy4 preview: Accuracy 71.7%, latency (mean per test) 2,764 s Qwen3.8 Max (0902): Accuracy 73.9%, latency (mean per test) 5,277 s Gemini 3.8 Flash: Accuracy 69.8%, latency (mean per test) 555 s GPT-6 Astra: Accuracy 68.4%, latency (mean per test) 1,448 s Step 5 Preview: Accuracy 70.2%, latency (mean per test) 1,483 s Grok 4.7: Accuracy 72.7%, latency (mean per test) 1,473 s MiMo-V2.6 Flash: Accuracy 69%, latency (mean per test) 1,100 s MiMo-V2.6 Pro: Accuracy 72.1%, latency (mean per test) 1,742 s Claude Opus 5.5: Accuracy 72.4%, latency (mean per test) 4,757 s GPT-6 Luna: Accuracy 62.2%, latency (mean per test) 1,950 s Claude Sonnet 5.5: Accuracy 68%, latency (mean per test) 4,516 s GPT-6.1 Sol: Accuracy 68.7%, latency (mean per test) 1,900 s Mistral Large 4: Accuracy 67.8%, latency (mean per test) 1,139 s Mistral Large 4 Preview: Accuracy 65.6%, latency (mean per test) 4,164 s Claude Haiku 5.5: Accuracy 70.3%, latency (mean per test) 3,525 s Muse Wonder: Accuracy 75.3%, latency (mean per test) 775 s Muse Spark Open: Accuracy 68.5%, latency (mean per test) 5,607 s Nemotron 3.5 Super VL: Accuracy 55.4%, latency (mean per test) 109 s DeepSeek V4.1 Flash 1017: Accuracy 71.9%, latency (mean per test) 1,567 s Mercury 3: Accuracy 63%, latency (mean per test) 534 s MiniMax-M3.1-Flash: Accuracy 67.1%, latency (mean per test) 2,124 s Kimi K3.1 Pro: Accuracy 67%, latency (mean per test) 4,710 s GPT-6.1 Terra: Accuracy 58.8%, latency (mean per test) 2,592 s Gemini 4 Carbon: Accuracy 75.9%, latency (mean per test) 2,307 s Tencent Hy4: Accuracy 70.5%, latency (mean per test) 3,477 s Inkling 2 Small: Accuracy 73.9%, latency (mean per test) 813 s Claude Fable 5.2: Accuracy 69.2%, latency (mean per test) 890 s Gemini 4 Hydrogen: Accuracy 73.6%, latency (mean per test) 1,267 s Gemini 4 Argon: Accuracy 76.8%, latency (mean per test) 1,541 s Beam 1.0: Accuracy 61.8%, latency (mean per test) 89.8 s Muse Wonder Full: Accuracy 77.9%, latency (mean per test) 1,796 s Muse Ember: Accuracy 75.7%, latency (mean per test) 217 s Muse Wonder FullGemini 4 ArgonMuse EmberBeam 1.0
ModelSettingAccuracyLatency (mean per test)
Muse Wonder Full 77.9% 1,796 s
Gemini 4 Argon 76.8% 1,541 s
Gemini 4 Carbon 75.9% 2,307 s
Muse Ember 75.7% 217.16 s
Muse Wonder 75.3% 775.37 s
GLM-5.3 74.6% 2,691 s
Qwen3.8 Max (0902) 73.9% 5,277 s
Inkling 2 Small 73.9% 813.38 s
Gemini 4 Hydrogen 73.6% 1,267 s
Grok 4.7 72.7% 1,473 s
Claude Opus 5.5 72.4% 4,757 s
MiMo-V2.6 Pro 72.1% 1,742 s
DeepSeek V4.1 Flash 1017 71.9% 1,567 s
Kimi K3 71.7% 1,911 s
Tencent Hy4 preview 71.7% 2,764 s
GLM-5.3-Flash 71.1% 1,703 s
Tencent Hy4 70.6% 3,477 s
Claude Haiku 5.5 70.3% 3,525 s
Step 5 Preview 70.2% 1,483 s
Gemini 3.8 Flash 69.8% 554.79 s
Claude Fable 5.2 69.2% 890.05 s
DeepSeek V4 Pro 0813 69% 1,593 s
MiMo-V2.6 Flash 69% 1,100 s
GPT-6.1 Sol 68.7% 1,900 s
Muse Spark Open 68.5% 5,607 s
GPT-6 Astra 68.4% 1,448 s
Claude Sonnet 5.5 68% 4,516 s
Mistral Large 4 67.8% 1,139 s
MiniMax-M3.1-Flash 67.1% 2,124 s
Kimi K3.1 Pro 67% 4,710 s
Mistral Large 4 Preview 65.6% 4,164 s
Mercury 3 63% 534.22 s
GPT-6 Luna 62.2% 1,950 s
Beam 1.0 61.8% 89.81 s
GPT-6.1 Terra 58.8% 2,592 s
Inkling-Small 57.8% 538.45 s
Nemotron 3.5 Super VL 55.4% 109.25 s
Inkling 54.2% 736.53 s

The best score, 77.9%, costs $3.72 a task

Runs on the cost frontier score higher than any run that's cheaper. Muse Ember comes within 2.2 points of the best for $0.07 a task, 53 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Meta Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Muse Wonder Full Muse Ember 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Muse Wonder Full

Source: Official Vals Legal Index leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Muse Wonder Full 77.9% $3.72 86,269
Gemini 4 Argon 76.8% $6.82 115,365
Gemini 4 Carbon 75.9% $7.20 191,000
Muse Ember 75.7% $0.07 62,734
Muse Wonder 75.3% $4.32 76,463
GLM-5.3 74.6% $2.68 215,835
Qwen3.8 Max (0902) 73.9% $4.25 250,336
Inkling 2 Small 73.9% $0.56 119,869
Gemini 4 Hydrogen 73.6% $4.47 311,609
Grok 4.7 72.7% $5.82 98,253
Claude Opus 5.5 72.4% $20 534,121
MiMo-V2.6 Pro 72.1% $0.18 84,137
DeepSeek V4.1 Flash 1017 71.9% $0.59 305,209
Kimi K3 71.7% $3.47 64,315
Tencent Hy4 preview 71.7% $0.81 164,728
GLM-5.3-Flash 71% $0.23 146,331
Tencent Hy4 70.5% $1.65 355,469
Claude Haiku 5.5 70.3% $2.36 849,224
Step 5 Preview 70.2% $0.84 106,185
Gemini 3.8 Flash 69.8% $2.05 81,067
Claude Fable 5.2 69.2% $8.88 91,879
DeepSeek V4 Pro 0813 69% $0.87 119,185
MiMo-V2.6 Flash 69% $0.07 85,085
GPT-6.1 Sol 68.7% $2.75 86,237
Muse Spark Open 68.5% $5.36 55,891
GPT-6 Astra 68.4% $14 84,047
Claude Sonnet 5.5 68% $13 674,634
Mistral Large 4 67.8% $0.78 76,745
MiniMax-M3.1-Flash 67.1% $0.41 340,898
Kimi K3.1 Pro 67% $6.63 223,400
Mistral Large 4 Preview 65.6% $2.22 142,245
Mercury 3 63% $0.10 78,229
GPT-6 Luna 62.2% $0.31 202,278
Beam 1.0 61.8% 27,621
GPT-6.1 Terra 58.8% $2.29 215,127
Inkling-Small 57.8% $0.14 45,477
Nemotron 3.5 Super VL 55.4% $0.16 37,024
Inkling 54.2% $0.70 48,454

2 models have held number one on the official leaderboard since its release in October 2026 2 models have held number one with claims included since its release in October 2026

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

Google DeepMind Meta
Own leaderboard 2 leaders
Muse Wonder Full 77.9%

Source: Official Vals Legal Index 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% 2027 Near the ceiling (85%) Muse Wonder Full GLM-5.3 Inkling-Small 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Muse Wonder Full Inkling-Small

Sources: Official Vals Legal Index 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.

Thinking Machines
DateLaunch postScoreAs the post writes it
Nov 5, 2026 Inkling 2 Small 73.9 Vals Legal Index

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
Inkling 2 Small 73.9% Vals Legal Index 73.9% Leaderboard Same

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%