Benchmark Almanac

Harvey LAB (Legal Agent Benchmark)

Muse Spark 1.2 25.4% leads since August 2026
Muse Spark 1.2 25.4% leads since August 2026
In use

Agent works through legal matter files across 24+ practice areas and produces work product. Expert rubric, task passes only if every criterion passes.

All-pass scoring keeps it far from saturated. Launch results used Harvey's private holdout. Vals HLAB v1.1 (October 21st, 2026) has 120 holdout tasks, and per-criterion pass is ~93-96%. Harvey ran its own baselines. Vals now runs the board. Repo keeps a score-impact CHANGELOG so results at two commits can be checked for comparability. Dataset field is a community Harbor port (1,760 tasks, not official).

Made by
Harvey
Released
May 2026
Kind
Law
Level
Expert
Scored
Graded by a model
Answers
Open-ended
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
80
Models scored on its own leaderboard
10
Launch posts list it, from 6 labs
11
System cards report it, from 3 labs

Muse Spark 1.2 leads with 25.4%

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.

Muse Spark 1.2 leads with 25.4%, 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.

80 models. Scroll sideways for the rest.

  1. 25.4 Muse Spark 1.2
  2. 23.8 Muse Spark 1.3
  3. 21.7 Muse Wonder Full
  4. 20.0 Muse Spark 1.1
  5. 19.6 Gemini 4 Argon
  6. 19.2 GLM Legal
  7. 17.1 Muse Wonder
  8. 16.7 Muse Spark Open
  9. 15.8 Grok 4.6
  10. 15.8 Mistral Large 4 Preview
  11. 12.9 Grok 4.5
  12. 12.9 Kimi K3
  13. 12.5 Grok 4.7
  14. 11.3 Claude Fable 5
  15. 11.3 Qwen 3.8 27B
  16. 11.3 MiMo-V2.6 Flash
  17. 10.8 MiMo-V2.6 Pro
  18. 10.4 Qwen 3.8 Max
  19. 10.0 Gemini 3.8 Flash
  20. 9.58 Claude Opus 4.8
  21. 9.17 Tencent Hy4 preview
  22. 9.17 Step 5 Preview
  23. 9.17 Ember-1
  24. 9.17 DeepSeek V4.1 Flash 1017
  25. 8.75 Gemini 3.7 Flash
  26. 8.33 DeepSeek V4 Flash 0731
  27. 8.33 GLM-5.3
  28. 7.50 DeepSeek V4 Pro 0813
  29. 7.08 GLM-5.2
  30. 6.67 Claude Opus 4.7
  31. 6.67 Claude Opus 5
  32. 6.67 GLM-5.3-Flash
  33. 6.67 Claude Fable 5.1
  34. 6.67 DeepSeek V4.1 Flash
  35. 5.42 GPT-6 Astra
  36. 5.42 GPT-6.1 Sol
  37. 5.00 Claude Sonnet 4.6
  38. 5.00 Claude Sonnet 5
  39. 4.17 MiniMax-M3
  40. 3.75 GPT-5.5
  41. 3.75 DeepSeek-V4-Pro
  42. 3.75 Claude Opus 5.5
  43. 3.33 Gemini 3.6 Flash
  44. 2.92 GPT-6 Luna
  45. 2.92 Claude Sonnet 5.5
  46. 2.50 Gemini 3.5 Flash
  47. 2.50 GPT-5.6 Sol
  48. 2.08 MiMo-V2.5-Pro
  49. 2.08 Inkling
  50. 1.67 Kimi K2.6
  51. 1.67 MiMo-V2.5
  52. 1.67 Qwen3.7-Max
  53. 1.67 Inkling-Small
  54. 1.67 GPT-6 Sol
  55. 1.25 Qwen 3.6 Plus
  56. 1.25 GPT-5.6 Luna
  57. 1.25 Ling-3.0-flash
  58. 1.25 Claude Haiku 5.5
  59. 0.83 Claude Haiku 4.5
  60. 0.83 GPT-5.6 Terra
  61. 0.42 Mistral Medium 3.5
  62. 0.42 Grok 4.3
  63. 0.42 Nemotron 3 Ultra
  64. 0.0 Gemini 3 Flash
  65. 0.0 Kimi K2.5
  66. 0.0 Grok 4.20
  67. 0.0 Gemini 3.1 Pro
  68. 0.0 Gemini 3.1 Flash-Lite
  69. 0.0 GPT-5.4
  70. 0.0 GPT-5.4 Mini
  71. 0.0 GPT-5.4 Nano
  72. 0.0 MiniMax-M2.7
  73. 0.0 GLM-5.1
  74. 0.0 Laguna M.1
  75. 0.0 Laguna XS.2
  76. 0.0 Qwen3.7-Plus
  77. 0.0 Gemini 3.5 Flash-Lite
  78. 0.0 Nemotron 3.5 Lightning 30B-A3B
  79. 0.0 Ling-3.0-flash-Fin
  80. 0.0 Mercury 2.5
  1. 1 Muse Spark 1.2 25.4%
  2. 2 Muse Spark 1.3 max 23.8%
  3. 3 Muse Wonder Full 21.7%
  4. 4 Muse Spark 1.1 20%
  5. 5 Gemini 4 Argon 19.6%
  6. 6 GLM Legal 19.2%
  7. 7 Muse Wonder 17.1%
  8. 8 Muse Spark Open 16.7%
  9. 9 Grok 4.6 15.8%
  10. 10 Mistral Large 4 Preview 15.8%
  11. 11 Grok 4.5 12.9%
  12. 12 Kimi K3 12.9%
  13. 13 Grok 4.7 12.5%
  14. 14 Claude Fable 5 11.3%
  15. 15 Qwen 3.8 27B 11.3%
  16. 16 MiMo-V2.6 Flash 11.3%
  17. 17 MiMo-V2.6 Pro 10.8%
  18. 18 Qwen 3.8 Max 10.4%
  19. 19 Gemini 3.8 Flash 10%
  20. 20 Claude Opus 4.8 9.58%
  21. 21 Tencent Hy4 preview 9.17%
  22. 22 Step 5 Preview 9.17%
  23. 23 Ember-1 9.17%
  24. 24 DeepSeek V4.1 Flash 1017 9.17%
  25. 25 Gemini 3.7 Flash 8.75%
  26. 26 DeepSeek V4 Flash 0731 8.33%
  27. 27 GLM-5.3 8.33%
  28. 28 DeepSeek V4 Pro 0813 7.50%
  29. 29 GLM-5.2 7.08%
  30. 30 Claude Opus 4.7 6.67%
  31. 31 Claude Opus 5 6.67%
  32. 32 GLM-5.3-Flash 6.67%
  33. 33 Claude Fable 5.1 6.67%
  34. 34 DeepSeek V4.1 Flash 6.67%
  35. 35 GPT-6 Astra 5.42%
  36. 36 GPT-6.1 Sol 5.42%
  37. 37 Claude Sonnet 4.6 5.00%
  38. 38 Claude Sonnet 5 5.00%
  39. 39 MiniMax-M3 4.17%
  40. 40 GPT-5.5 3.75%
  41. 41 DeepSeek-V4-Pro 3.75%
  42. 42 Claude Opus 5.5 3.75%
  43. 43 Gemini 3.6 Flash 3.33%
  44. 44 GPT-6 Luna 2.92%
  45. 45 Claude Sonnet 5.5 2.92%
  46. 46 Gemini 3.5 Flash 2.50%
  47. 47 GPT-5.6 Sol 2.50%
  48. 48 MiMo-V2.5-Pro 2.08%
  49. 49 Inkling 2.08%
  50. 50 Kimi K2.6 1.67%
  51. 51 MiMo-V2.5 1.67%
  52. 52 Qwen3.7-Max 1.67%
  53. 53 Inkling-Small 1.67%
  54. 54 GPT-6 Sol 1.67%
  55. 55 Qwen 3.6 Plus 1.25%
  56. 56 GPT-5.6 Luna 1.25%
  57. 57 Ling-3.0-flash 1.25%
  58. 58 Claude Haiku 5.5 1.25%
  59. 59 Claude Haiku 4.5 thinking 0.83%
  60. 60 GPT-5.6 Terra 0.83%
  61. 61 Mistral Medium 3.5 0.42%
  62. 62 Grok 4.3 0.42%
  63. 63 Nemotron 3 Ultra 0.42%
  64. 64 Gemini 3 Flash 0%
  65. 65 Kimi K2.5 thinking 0%
  66. 66 Grok 4.20 reasoning 0%
  67. 67 Gemini 3.1 Pro 0%
  68. 68 Gemini 3.1 Flash-Lite 0%
  69. 69 GPT-5.4 xhigh 0%
  70. 70 GPT-5.4 Mini 0%
  71. 71 GPT-5.4 Nano 0%
  72. 72 MiniMax-M2.7 0%
  73. 73 GLM-5.1 0%
  74. 74 Laguna M.1 0%
  75. 75 Laguna XS.2 0%
  76. 76 Qwen3.7-Plus 0%
  77. 77 Gemini 3.5 Flash-Lite 0%
  78. 78 Nemotron 3.5 Lightning 30B-A3B 0%
  79. 79 Ling-3.0-flash-Fin 0%
  80. 80 Mercury 2.5 0%

Source: Official Harvey LAB (Legal Agent Benchmark) leaderboard

83 models. Scroll sideways for the rest.

  1. 25.4 Muse Spark 1.2
  2. 23.8 Muse Spark 1.3
  3. 23.7 Kimi K3.1 Pro
  4. 23.6 Claude Opus 5
  5. 21.7 Muse Wonder Full
  6. 20.0 Muse Spark 1.1
  7. 19.6 Grok 4.7
  8. 19.6 Gemini 4 Argon
  9. 19.2 GLM-5.3-FlashX-Legal
  10. 19.2 GLM Legal
  11. 19.1 Claude Fable 5.1
  12. 17.1 Muse Wonder
  13. 16.9 Claude Fable 5
  14. 16.7 Muse Spark Open
  15. 15.8 Grok 4.6
  16. 15.8 Mistral Large 4 Preview
  17. 15.8 Mistral Large 4
  18. 12.9 Grok 4.5
  19. 12.9 Kimi K3
  20. 11.7 Claude Sonnet 5.5
  21. 11.3 Qwen 3.8 27B
  22. 11.3 MiMo-V2.6 Flash
  23. 10.8 MiMo-V2.6 Pro
  24. 10.4 Qwen 3.8 Max
  25. 10.0 Gemini 3.8 Flash
  26. 9.62 Claude Opus 4.8
  27. 9.17 Tencent Hy4 preview
  28. 9.17 Step 5 Preview
  29. 9.17 Ember-1
  30. 9.17 DeepSeek V4.1 Flash 1017
  31. 8.90 Claude Sonnet 5
  32. 8.75 Gemini 3.7 Flash
  33. 8.33 DeepSeek V4 Flash 0731
  34. 8.33 GLM-5.3
  35. 8.30 Claude Opus 5.5
  36. 7.50 DeepSeek V4 Pro 0813
  37. 7.08 GLM-5.2
  38. 6.67 Claude Opus 4.7
  39. 6.67 GLM-5.3-Flash
  40. 6.67 DeepSeek V4.1 Flash
  41. 5.42 GPT-6 Astra
  42. 5.42 GPT-6.1 Sol
  43. 5.00 Claude Sonnet 4.6
  44. 4.17 MiniMax-M3
  45. 3.75 GPT-5.5
  46. 3.75 DeepSeek-V4-Pro
  47. 3.33 Gemini 3.6 Flash
  48. 2.92 GPT-6 Luna
  49. 2.50 Gemini 3.5 Flash
  50. 2.50 GPT-5.6 Sol
  51. 2.08 MiMo-V2.5-Pro
  52. 2.08 Inkling
  53. 1.67 Kimi K2.6
  54. 1.67 MiMo-V2.5
  55. 1.67 Qwen3.7-Max
  56. 1.67 Inkling-Small
  57. 1.67 GPT-6 Sol
  58. 1.25 Qwen 3.6 Plus
  59. 1.25 GPT-5.6 Luna
  60. 1.25 Ling-3.0-flash
  61. 1.25 Claude Haiku 5.5
  62. 0.83 Claude Haiku 4.5
  63. 0.83 GPT-5.6 Terra
  64. 0.42 Mistral Medium 3.5
  65. 0.42 Grok 4.3
  66. 0.42 Nemotron 3 Ultra
  67. 0.0 Gemini 3 Flash
  68. 0.0 Kimi K2.5
  69. 0.0 Grok 4.20
  70. 0.0 Gemini 3.1 Pro
  71. 0.0 Gemini 3.1 Flash-Lite
  72. 0.0 GPT-5.4
  73. 0.0 GPT-5.4 Mini
  74. 0.0 GPT-5.4 Nano
  75. 0.0 MiniMax-M2.7
  76. 0.0 GLM-5.1
  77. 0.0 Laguna M.1
  78. 0.0 Laguna XS.2
  79. 0.0 Qwen3.7-Plus
  80. 0.0 Gemini 3.5 Flash-Lite
  81. 0.0 Nemotron 3.5 Lightning 30B-A3B
  82. 0.0 Ling-3.0-flash-Fin
  83. 0.0 Mercury 2.5
  1. 1 Muse Spark 1.2 25.4%
  2. 2 Muse Spark 1.3 max 23.8%
  3. 3 Kimi K3.1 Pro 23.7%
  4. 4 Claude Opus 5 23.6%
  5. 5 Muse Wonder Full 21.7%
  6. 6 Muse Spark 1.1 20%
  7. 7 Grok 4.7 19.6%
  8. 8 Gemini 4 Argon 19.6%
  9. 9 GLM-5.3-FlashX-Legal 19.2%
  10. 10 GLM Legal 19.2%
  11. 11 Claude Fable 5.1 19.1%
  12. 12 Muse Wonder 17.1%
  13. 13 Claude Fable 5 16.9%
  14. 14 Muse Spark Open 16.7%
  15. 15 Grok 4.6 15.8%
  16. 16 Mistral Large 4 Preview 15.8%
  17. 17 Mistral Large 4 15.8%
  18. 18 Grok 4.5 12.9%
  19. 19 Kimi K3 12.9%
  20. 20 Claude Sonnet 5.5 11.7%
  21. 21 Qwen 3.8 27B 11.3%
  22. 22 MiMo-V2.6 Flash 11.3%
  23. 23 MiMo-V2.6 Pro 10.8%
  24. 24 Qwen 3.8 Max 10.4%
  25. 25 Gemini 3.8 Flash 10%
  26. 26 Claude Opus 4.8 9.62%
  27. 27 Tencent Hy4 preview 9.17%
  28. 28 Step 5 Preview 9.17%
  29. 29 Ember-1 9.17%
  30. 30 DeepSeek V4.1 Flash 1017 9.17%
  31. 31 Claude Sonnet 5 8.90%
  32. 32 Gemini 3.7 Flash 8.75%
  33. 33 DeepSeek V4 Flash 0731 8.33%
  34. 34 GLM-5.3 8.33%
  35. 35 Claude Opus 5.5 8.30%
  36. 36 DeepSeek V4 Pro 0813 7.50%
  37. 37 GLM-5.2 7.08%
  38. 38 Claude Opus 4.7 6.67%
  39. 39 GLM-5.3-Flash 6.67%
  40. 40 DeepSeek V4.1 Flash 6.67%
  41. 41 GPT-6 Astra 5.42%
  42. 42 GPT-6.1 Sol 5.42%
  43. 43 Claude Sonnet 4.6 5.00%
  44. 44 MiniMax-M3 4.17%
  45. 45 GPT-5.5 3.75%
  46. 46 DeepSeek-V4-Pro 3.75%
  47. 47 Gemini 3.6 Flash 3.33%
  48. 48 GPT-6 Luna 2.92%
  49. 49 Gemini 3.5 Flash 2.50%
  50. 50 GPT-5.6 Sol 2.50%
  51. 51 MiMo-V2.5-Pro 2.08%
  52. 52 Inkling 2.08%
  53. 53 Kimi K2.6 1.67%
  54. 54 MiMo-V2.5 1.67%
  55. 55 Qwen3.7-Max 1.67%
  56. 56 Inkling-Small 1.67%
  57. 57 GPT-6 Sol 1.67%
  58. 58 Qwen 3.6 Plus 1.25%
  59. 59 GPT-5.6 Luna 1.25%
  60. 60 Ling-3.0-flash 1.25%
  61. 61 Claude Haiku 5.5 1.25%
  62. 62 Claude Haiku 4.5 thinking 0.83%
  63. 63 GPT-5.6 Terra 0.83%
  64. 64 Mistral Medium 3.5 0.42%
  65. 65 Grok 4.3 0.42%
  66. 66 Nemotron 3 Ultra 0.42%
  67. 67 Gemini 3 Flash 0%
  68. 68 Kimi K2.5 thinking 0%
  69. 69 Grok 4.20 reasoning 0%
  70. 70 Gemini 3.1 Pro 0%
  71. 71 Gemini 3.1 Flash-Lite 0%
  72. 72 GPT-5.4 xhigh 0%
  73. 73 GPT-5.4 Mini 0%
  74. 74 GPT-5.4 Nano 0%
  75. 75 MiniMax-M2.7 0%
  76. 76 GLM-5.1 0%
  77. 77 Laguna M.1 0%
  78. 78 Laguna XS.2 0%
  79. 79 Qwen3.7-Plus 0%
  80. 80 Gemini 3.5 Flash-Lite 0%
  81. 81 Nemotron 3.5 Lightning 30B-A3B 0%
  82. 82 Ling-3.0-flash-Fin 0%
  83. 83 Mercury 2.5 0%

Sources: Official Harvey LAB (Legal Agent Benchmark) leaderboard and labs’ model cards

#ModelScoreSettingDateSource
1 Muse Spark 1.2 25.4% Aug 2026 Official leaderboard
2 Muse Spark 1.3 23.8% max Sep 2026 Official leaderboard
3 Muse Wonder Full 21.7% Oct 2026 Official leaderboard
4 Muse Spark 1.1 20% Jul 2026 Official leaderboard
5 Gemini 4 Argon 19.6% Sep 2026 Official leaderboard
6 GLM Legal 19.2% Oct 2026 Official leaderboard
7 Muse Wonder 17.1% Oct 2026 Official leaderboard
8 Muse Spark Open 16.7% Oct 2026 Official leaderboard
9 Grok 4.6 15.8% Aug 2026 Official leaderboard
10 Mistral Large 4 Preview 15.8% Oct 2026 Official leaderboard
11 Grok 4.5 12.9% Jul 2026 Official leaderboard
12 Kimi K3 12.9% Jul 2026 Official leaderboard
13 Grok 4.7 12.5% Sep 2026 Official leaderboard
14 Claude Fable 5 11.3% Jun 2026 Official leaderboard
15 Qwen 3.8 27B 11.3% Aug 2026 Official leaderboard
16 MiMo-V2.6 Flash 11.3% Sep 2026 Official leaderboard
17 MiMo-V2.6 Pro 10.8% Sep 2026 Official leaderboard
18 Qwen 3.8 Max 10.4% Aug 2026 Official leaderboard
19 Gemini 3.8 Flash 10% Sep 2026 Official leaderboard
20 Claude Opus 4.8 9.58% May 2026 Official leaderboard
21 Tencent Hy4 preview 9.17% Aug 2026 Official leaderboard
22 Step 5 Preview 9.17% Sep 2026 Official leaderboard
23 Ember-1 9.17% Sep 2026 Official leaderboard
24 DeepSeek V4.1 Flash 1017 9.17% Oct 2026 Official leaderboard
25 Gemini 3.7 Flash 8.75% Aug 2026 Official leaderboard
26 DeepSeek V4 Flash 0731 8.33% Jul 2026 Official leaderboard
27 GLM-5.3 8.33% Aug 2026 Official leaderboard
28 DeepSeek V4 Pro 0813 7.50% Aug 2026 Official leaderboard
29 GLM-5.2 7.08% Jun 2026 Official leaderboard
30 Claude Opus 4.7 6.67% Apr 2026 Official leaderboard
31 Claude Opus 5 6.67% Jul 2026 Official leaderboard
32 GLM-5.3-Flash 6.67% Aug 2026 Official leaderboard
33 Claude Fable 5.1 6.67% Sep 2026 Official leaderboard
34 DeepSeek V4.1 Flash 6.67% Sep 2026 Official leaderboard
35 GPT-6 Astra 5.42% Sep 2026 Official leaderboard
36 GPT-6.1 Sol 5.42% Sep 2026 Official leaderboard
37 Claude Sonnet 4.6 5.00% Feb 2026 Official leaderboard
38 Claude Sonnet 5 5.00% Jun 2026 Official leaderboard
39 MiniMax-M3 4.17% Jun 2026 Official leaderboard
40 GPT-5.5 3.75% Apr 2026 Official leaderboard
41 DeepSeek-V4-Pro 3.75% Apr 2026 Official leaderboard
42 Claude Opus 5.5 3.75% Sep 2026 Official leaderboard
43 Gemini 3.6 Flash 3.33% Jul 2026 Official leaderboard
44 GPT-6 Luna 2.92% Sep 2026 Official leaderboard
45 Claude Sonnet 5.5 2.92% Sep 2026 Official leaderboard
46 Gemini 3.5 Flash 2.50% May 2026 Official leaderboard
47 GPT-5.6 Sol 2.50% Jul 2026 Official leaderboard
48 MiMo-V2.5-Pro 2.08% Apr 2026 Official leaderboard
49 Inkling 2.08% Jul 2026 Official leaderboard
50 Kimi K2.6 1.67% Apr 2026 Official leaderboard
51 MiMo-V2.5 1.67% Apr 2026 Official leaderboard
52 Qwen3.7-Max 1.67% May 2026 Official leaderboard
53 Inkling-Small 1.67% Jul 2026 Official leaderboard
54 GPT-6 Sol 1.67% Sep 2026 Official leaderboard
55 Qwen 3.6 Plus 1.25% Mar 2026 Official leaderboard
56 GPT-5.6 Luna 1.25% Jul 2026 Official leaderboard
57 Ling-3.0-flash 1.25% Aug 2026 Official leaderboard
58 Claude Haiku 5.5 1.25% Oct 2026 Official leaderboard
59 Claude Haiku 4.5 0.83% thinking Oct 2025 Official leaderboard
60 GPT-5.6 Terra 0.83% Jul 2026 Official leaderboard
61 Mistral Medium 3.5 0.42% Apr 2026 Official leaderboard
62 Grok 4.3 0.42% Apr 2026 Official leaderboard
63 Nemotron 3 Ultra 0.42% Jun 2026 Official leaderboard
64 Gemini 3 Flash 0% Dec 2025 Official leaderboard
65 Kimi K2.5 0% thinking Jan 2026 Official leaderboard
66 Grok 4.20 0% reasoning Feb 2026 Official leaderboard
67 Gemini 3.1 Pro 0% Feb 2026 Official leaderboard
68 Gemini 3.1 Flash-Lite 0% Mar 2026 Official leaderboard
69 GPT-5.4 0% xhigh Mar 2026 Official leaderboard
70 GPT-5.4 Mini 0% Mar 2026 Official leaderboard
71 GPT-5.4 Nano 0% Mar 2026 Official leaderboard
72 MiniMax-M2.7 0% Mar 2026 Official leaderboard
73 GLM-5.1 0% Apr 2026 Official leaderboard
74 Laguna M.1 0% Apr 2026 Official leaderboard
75 Laguna XS.2 0% Apr 2026 Official leaderboard
76 Qwen3.7-Plus 0% Jun 2026 Official leaderboard
77 Gemini 3.5 Flash-Lite 0% Jul 2026 Official leaderboard
78 Nemotron 3.5 Lightning 30B-A3B 0% Aug 2026 Official leaderboard
79 Ling-3.0-flash-Fin 0% Sep 2026 Official leaderboard
80 Mercury 2.5 0% Sep 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Muse Spark 1.2 25.4% Aug 2026 Official leaderboard
2 Muse Spark 1.3 23.8% max Sep 2026 Official leaderboard
3 Kimi K3.1 Pro 23.7% Oct 2026 Lab model card
4 Claude Opus 5 23.6% Jul 2026 Lab model card
5 Muse Wonder Full 21.7% Oct 2026 Official leaderboard
6 Muse Spark 1.1 20% Jul 2026 Official leaderboard
7 Grok 4.7 19.6% Sep 2026 Lab model card
8 Gemini 4 Argon 19.6% Oct 2026 Lab model card
9 GLM-5.3-FlashX-Legal 19.2% Oct 2026 Lab model card
10 GLM Legal 19.2% Oct 2026 Official leaderboard
11 Claude Fable 5.1 19.1% Sep 2026 Lab model card
12 Muse Wonder 17.1% Oct 2026 Official leaderboard
13 Claude Fable 5 16.9% Jun 2026 Lab model card
14 Muse Spark Open 16.7% Oct 2026 Official leaderboard
15 Grok 4.6 15.8% Aug 2026 Official leaderboard
16 Mistral Large 4 Preview 15.8% Oct 2026 Official leaderboard
17 Mistral Large 4 15.8% Oct 2026 Lab model card
18 Grok 4.5 12.9% Jul 2026 Official leaderboard
19 Kimi K3 12.9% Jul 2026 Official leaderboard
20 Claude Sonnet 5.5 11.7% Sep 2026 Lab model card
21 Qwen 3.8 27B 11.3% Aug 2026 Official leaderboard
22 MiMo-V2.6 Flash 11.3% Sep 2026 Official leaderboard
23 MiMo-V2.6 Pro 10.8% Sep 2026 Official leaderboard
24 Qwen 3.8 Max 10.4% Aug 2026 Official leaderboard
25 Gemini 3.8 Flash 10% Sep 2026 Official leaderboard
26 Claude Opus 4.8 9.62% May 2026 Lab model card
27 Tencent Hy4 preview 9.17% Aug 2026 Official leaderboard
28 Step 5 Preview 9.17% Sep 2026 Official leaderboard
29 Ember-1 9.17% Sep 2026 Official leaderboard
30 DeepSeek V4.1 Flash 1017 9.17% Oct 2026 Official leaderboard
31 Claude Sonnet 5 8.90% Jun 2026 Lab model card
32 Gemini 3.7 Flash 8.75% Aug 2026 Official leaderboard
33 DeepSeek V4 Flash 0731 8.33% Jul 2026 Official leaderboard
34 GLM-5.3 8.33% Aug 2026 Official leaderboard
35 Claude Opus 5.5 8.30% Sep 2026 Lab model card
36 DeepSeek V4 Pro 0813 7.50% Aug 2026 Official leaderboard
37 GLM-5.2 7.08% Jun 2026 Official leaderboard
38 Claude Opus 4.7 6.67% Apr 2026 Official leaderboard
39 GLM-5.3-Flash 6.67% Aug 2026 Official leaderboard
40 DeepSeek V4.1 Flash 6.67% Sep 2026 Official leaderboard
41 GPT-6 Astra 5.42% Sep 2026 Official leaderboard
42 GPT-6.1 Sol 5.42% Sep 2026 Official leaderboard
43 Claude Sonnet 4.6 5.00% Feb 2026 Official leaderboard
44 MiniMax-M3 4.17% Jun 2026 Official leaderboard
45 GPT-5.5 3.75% Apr 2026 Official leaderboard
46 DeepSeek-V4-Pro 3.75% Apr 2026 Official leaderboard
47 Gemini 3.6 Flash 3.33% Jul 2026 Official leaderboard
48 GPT-6 Luna 2.92% Sep 2026 Official leaderboard
49 Gemini 3.5 Flash 2.50% May 2026 Official leaderboard
50 GPT-5.6 Sol 2.50% Jul 2026 Official leaderboard
51 MiMo-V2.5-Pro 2.08% Apr 2026 Official leaderboard
52 Inkling 2.08% Jul 2026 Official leaderboard
53 Kimi K2.6 1.67% Apr 2026 Official leaderboard
54 MiMo-V2.5 1.67% Apr 2026 Official leaderboard
55 Qwen3.7-Max 1.67% May 2026 Official leaderboard
56 Inkling-Small 1.67% Jul 2026 Official leaderboard
57 GPT-6 Sol 1.67% Sep 2026 Official leaderboard
58 Qwen 3.6 Plus 1.25% Mar 2026 Official leaderboard
59 GPT-5.6 Luna 1.25% Jul 2026 Official leaderboard
60 Ling-3.0-flash 1.25% Aug 2026 Official leaderboard
61 Claude Haiku 5.5 1.25% Oct 2026 Official leaderboard
62 Claude Haiku 4.5 0.83% thinking Oct 2025 Official leaderboard
63 GPT-5.6 Terra 0.83% Jul 2026 Official leaderboard
64 Mistral Medium 3.5 0.42% Apr 2026 Official leaderboard
65 Grok 4.3 0.42% Apr 2026 Official leaderboard
66 Nemotron 3 Ultra 0.42% Jun 2026 Official leaderboard
67 Gemini 3 Flash 0% Dec 2025 Official leaderboard
68 Kimi K2.5 0% thinking Jan 2026 Official leaderboard
69 Grok 4.20 0% reasoning Feb 2026 Official leaderboard
70 Gemini 3.1 Pro 0% Feb 2026 Official leaderboard
71 Gemini 3.1 Flash-Lite 0% Mar 2026 Official leaderboard
72 GPT-5.4 0% xhigh Mar 2026 Official leaderboard
73 GPT-5.4 Mini 0% Mar 2026 Official leaderboard
74 GPT-5.4 Nano 0% Mar 2026 Official leaderboard
75 MiniMax-M2.7 0% Mar 2026 Official leaderboard
76 GLM-5.1 0% Apr 2026 Official leaderboard
77 Laguna M.1 0% Apr 2026 Official leaderboard
78 Laguna XS.2 0% Apr 2026 Official leaderboard
79 Qwen3.7-Plus 0% Jun 2026 Official leaderboard
80 Gemini 3.5 Flash-Lite 0% Jul 2026 Official leaderboard
81 Nemotron 3.5 Lightning 30B-A3B 0% Aug 2026 Official leaderboard
82 Ling-3.0-flash-Fin 0% Sep 2026 Official leaderboard
83 Mercury 2.5 0% Sep 2026 Official leaderboard

Labs report scores for 14 models in their cards. Switch the scores above to rank them together.

The best score, 25.4%, costs $2.09 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 Meta DeepSeek NVIDIA Z.ai Cost frontier One model at each effort level
0% 10% 20% 30% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Muse Spark 1.2 Muse Spark 1.1 GLM Legal MiMo-V2.6 Flash DeepSeek V4 Flash 0731 MiMo-V2.5 Nemotron 3.5 Lightning 30B-A3B 0% 10% 20% 30% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Muse Spark 1.2 Muse Spark 1.1

Source: Official Harvey LAB (Legal Agent Benchmark) leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Muse Spark 1.2 25.4% $2.09
Muse Spark 1.3 max 23.8% $2.26
Muse Spark 1.3 22.9% $2.76
Muse Wonder Full 21.7% $5.41
Muse Spark 1.1 20% $0.80
Gemini 4 Argon 19.6% $11
GLM Legal 19.2% $0.15
Muse Wonder 17.1% $7.08
Muse Spark Open 16.7% $6.83
Grok 4.6 15.8% $4.01
Mistral Large 4 Preview 15.8% $4.18
Grok 4.5 12.9% $2.02
Kimi K3 12.9% $3.83
Grok 4.7 12.5% $11
Claude Fable 5 11.3% $19
Qwen 3.8 27B 11.3% $4.03
MiMo-V2.6 Flash 11.3% $0.09
MiMo-V2.6 Pro 10.8% $0.22
Qwen 3.8 Max 10.4% $2.37
Gemini 3.8 Flash 10% $3.66
Claude Opus 4.8 9.58% $10
Tencent Hy4 preview 9.17% $0.98
Step 5 Preview 9.17% $1.00
Ember-1 9.17% $3.33
DeepSeek V4.1 Flash 1017 9.17% $0.54
Gemini 3.7 Flash 8.75% $2.56
DeepSeek V4 Flash 0731 8.33% $0.06
GLM-5.3 8.33% $4.30
DeepSeek V4 Pro 0813 7.50% $0.17
GLM-5.2 7.08% $2.06
Claude Opus 4.7 6.67% $11
Claude Opus 5 6.67% $24
GLM-5.3-Flash 6.67% $0.57
Claude Fable 5.1 6.67% $46
DeepSeek V4.1 Flash 6.67% $0.17
GPT-6 Astra 5.42% $26
GPT-6.1 Sol 5.42% $3.93
Claude Sonnet 4.6 5.00% $3.04
Claude Sonnet 5 5.00% $8.95
MiniMax-M3 4.17% $1.46
GPT-5.5 3.75% $4.60
DeepSeek-V4-Pro 3.75% $0.74
Claude Opus 5.5 3.75% $21
Gemini 3.6 Flash 3.33% $1.90
GPT-6 Luna 2.92% $0.30
Claude Sonnet 5.5 2.92% $17
Gemini 3.5 Flash 2.50% $2.12
GPT-5.6 Sol 2.50% $10
MiMo-V2.5-Pro 2.08% $0.11
Inkling 2.08% $1.25
Kimi K2.6 1.67% $0.82
MiMo-V2.5 1.67% $0.04
Qwen3.7-Max 1.67% $1.58
Inkling-Small 1.67% $0.17
GPT-6 Sol 1.67% $3.36
Qwen 3.6 Plus 1.25% $3.94
GPT-5.6 Luna 1.25% $0.38
Ling-3.0-flash 1.25% $0.05
Claude Haiku 5.5 1.25% $3.39
Claude Haiku 4.5 thinking 0.83% $0.42
GPT-5.6 Terra 0.83% $3.20
Mistral Medium 3.5 0.42% $6.07
Grok 4.3 0.42% $0.39
Nemotron 3 Ultra 0.42%
Gemini 3 Flash 0% $0.26
Kimi K2.5 thinking 0% $0.35
Grok 4.20 reasoning 0% $0.99
Gemini 3.1 Pro 0% $1.15
Gemini 3.1 Flash-Lite 0% $0.08
GPT-5.4 xhigh 0% $3.60
GPT-5.4 Mini 0% $1.00
GPT-5.4 Nano 0% $0.18
MiniMax-M2.7 0% $0.49
GLM-5.1 0% $0.56
Laguna M.1 0%
Laguna XS.2 0%
Qwen3.7-Plus 0% $0.23
Gemini 3.5 Flash-Lite 0% $0.33
Nemotron 3.5 Lightning 30B-A3B 0% $0.03
Ling-3.0-flash-Fin 0% $0.05
Mercury 2.5 0% $0.19

6 models have held number one on the official leaderboard since its release in May 2026 6 models have held number one with claims included since its release in May 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Anthropic xAI Meta Moonshot AI
Own leaderboard 6 leaders
Muse Spark 1.2 25.4%
Lab claims 4 leaders
Kimi K3.1 Pro 23.7%

Sources: Official Harvey LAB (Legal Agent Benchmark) leaderboard and labs’ model cards

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 2027 Near the ceiling (85%) Muse Spark 1.2 Muse Spark 1.1 Claude Opus 4.8 Claude Sonnet 4.6 Claude Haiku 4.5 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Muse Spark 1.2 Claude Fable 5

Sources: Official Harvey LAB (Legal Agent Benchmark) leaderboard and labs’ model cards

6 labs 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.

Anthropic
xAI
Google DeepMind
Mistral AI
Z.ai
Moonshot AI
DateLaunch postScoreAs the post writes it
Oct 24, 2026 Kimi K3.1 Pro 23.7 Harvey Legal Agent (public 1,251 tasks, Stirrup harness, all-pass rate)
Oct 13, 2026 Gemini 4 Argon 19.6 Harvey's Legal Agent Benchmark
Oct 12, 2026 GLM-5.3-FlashX-Legal 19.2 Harvey LAB all-criteria pass rate 19.2%
Oct 6, 2026 Mistral Large 4 15.8 Harvey's Legal Agent: Vals.ai - Harvey's Legal Agent Benchmark (Harvey LAB)
Sep 21, 2026 Grok 4.7 19.6 Harvey Legal Agent Benchmark (Legal work)
Sep 2, 2026 Gemini 3.8 Flash 10 Harvey's Legal Agent Benchmark (All pass rate)
Aug 12, 2026 Grok 4.6 15.8 Harvey LAB (Vals)
Jul 24, 2026 Claude Opus 5 11.7 Legal Legal Agent Benchmark, Held-out
Jun 30, 2026 Claude Sonnet 5 8.9 Best of Legal Agent Benchmark Full Public Set / Legal Agent Benchmark Harvey's Held-Out Set
Jun 9, 2026 Claude Fable 5 16.9 Best of Legal Legal Agent Benchmark (Harvey's Held-Out Set) / Legal Agent Full Public Benchmark Set

11 system cards report it, from 3 labs

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 19.6 Harvey's Legal Agent Benchmark (results table, p.5) Capabilities
Sep 28, 2026 Claude Sonnet 5.5 system card 11.7 Legal Agent Benchmark held-out set, all-pass rate at high effort, mean criterion pass 92.1% (max effort 10% and 93.1%) (8.14.2) Capabilities
Sep 22, 2026 Claude Opus 5.5 system card 8.3 Legal Agent Benchmark (Harvey LAB) held-out set of 120 problems, all-pass rate at max effort, run by Artificial Analysis (mean criterion pass rate 91.2%) (8.14.2) Capabilities
Sep 21, 2026 Grok 4.7 model card 19.6 Legal Agent Benchmark, Harvey final score, xhigh, Vals AI 120-task held-out run in Valkyrie (3.1) Capabilities
Sep 2, 2026 Gemini 3.8 Flash model card 10 Harvey's Legal Agent Benchmark, all pass rate (results table, p.5) Capabilities
Sep 1, 2026 Claude Fable 5.1 system card 19.1 Legal Agent Benchmark (Harvey LAB), all-pass rate on 1,235 public problems, max effort, internal harness, mean criterion pass rate 90.81%; held-out set run by Artificial Analysis 16.7% all-pass at xhigh (8.15.2) Capabilities
Aug 12, 2026 Grok 4.6 model card 15.8 Legal Agent Benchmark (LAB), score, high, Vals AI Valkyrie harness (3.5) Capabilities
Jul 24, 2026 Claude Opus 5 system card 23.6 Legal Agent Benchmark (Harvey LAB), all-pass rate on 1,235 public problems, max effort, internal harness, production safeguards, mean criterion pass rate 93.74%. Harvey's held-out set: 11.7% all-pass, 94.1% mean criterion (8.13.3) Capabilities
Jun 30, 2026 Claude Sonnet 5 system card 5.8 Legal Agent Benchmark, Harvey's held-out set, all-pass rate per Harvey's evaluation, mean criterion pass 91.2%. Full public set (1,235 problems, internal harness) 8.92% all-pass and 88.26% mean criterion (Table 8.1.A, 8.11.3) Capabilities
Jun 9, 2026 Claude Fable 5 system card 13.3 Legal Agent Benchmark, Harvey's held-out set, all-pass rate, Harvey's evaluation as of June 2026 (Table 8.1.A, 8.17.4) Capabilities
May 28, 2026 Claude Opus 4.8 system card 9.62 Legal Agent Benchmark (Harvey LAB), all-pass rate on 1,235 public problems, internal reimplementation, max effort, five trials, mean criterion pass rate 89.01% (8.13.3) Capabilities

The lab's number comes out ahead in 7 of 7 pairs that differ

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. 4 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap What differs
Claude Opus 5 23.6% Legal Agent Benchmark (Harvey LAB), all-pass rate on 1,235 public problems, max effort, internal harness, production safeguards, mean criterion pass rate 93.74%. Harvey's held-out set: 11.7% all-pass, 94.1% mean criterion (8.13.3) 6.67% Leaderboard +17 pts Anthropic evaluates 1,235 public problems, while Vals evaluates Harvey's private holdout with its benchmark harness.
Claude Fable 5.1 19.1% Legal Agent Benchmark (Harvey LAB), all-pass rate on 1,235 public problems, max effort, internal harness, mean criterion pass rate 90.81%; held-out set run by Artificial Analysis 16.7% all-pass at xhigh (8.15.2) 6.67% Leaderboard +12 pts
Claude Sonnet 5.5 11.7% Legal Agent Benchmark held-out set, all-pass rate at high effort, mean criterion pass 92.1% (max effort 10% and 93.1%) (8.14.2) 2.92% Leaderboard +8.8 pts
Grok 4.7 19.6% Harvey Legal Agent Benchmark (Legal work) 12.5% Leaderboard +7.1 pts
Claude Fable 5 16.9% Best of Legal Legal Agent Benchmark (Harvey's Held-Out Set) / Legal Agent Full Public Benchmark Set 11.3% Leaderboard +5.6 pts
Claude Opus 5.5 8.30% Legal Agent Benchmark (Harvey LAB) held-out set of 120 problems, all-pass rate at max effort, run by Artificial Analysis (mean criterion pass rate 91.2%) (8.14.2) 3.75% Leaderboard +4.6 pts
Claude Sonnet 5 8.90% Best of Legal Agent Benchmark Full Public Set / Legal Agent Benchmark Harvey's Held-Out Set 5.00% Leaderboard +3.9 pts
Claude Opus 4.8 9.62% Legal Agent Benchmark (Harvey LAB), all-pass rate on 1,235 public problems, internal reimplementation, max effort, five trials, mean criterion pass rate 89.01% (8.13.3) 9.58% Leaderboard Same
Grok 4.6 15.8% Harvey LAB (Vals) 15.8% Leaderboard Same
Gemini 4 Argon 19.6% Harvey's Legal Agent Benchmark 19.6% Leaderboard Same
Gemini 3.8 Flash 10% Harvey's Legal Agent Benchmark (All pass rate) 10% Leaderboard Same

2 logged changes that move scores

Changes by the maintainers, so a score from before a change may not compare with one from after it. Newest first, each with the maintainers' own source. The list may be incomplete. What we log.

  1. Tasks changed August 7th, 2026

    Harvey added a firm-knowledge environment with 250 tasks. Runs including this expansion test retrieval across a much larger document collection. Source

  2. Tasks changed June 12th, 2026

    Harvey added 500 contracting tasks to LAB. Results covering the expanded benchmark measure a broader task mix than the original law-firm release. Source

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 xAI Meta Google DeepMind
Harvey LAB v1.1 (HLAB-Vals-1.1) 7 leaders
Gemini 4 Argon 27.1%
Harvey LAB (Legal Agent Benchmark) 6 leaders
Muse Spark 1.2 25.4%
Harvey LAB v1.1 (HLAB-Vals-1.1) 1 leader
Gemini 4 Argon 27.1%

Source: Official leaderboards

Replaced by Harvey LAB v1.1 (HLAB-Vals-1.1)

More in Law, finance and medicine

All 179 in Law, finance and medicine
  • MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy. Qwen 3.8 Max 80.4%
  • HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
  • HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy. Ling-3.1-flash 81.7%
  • KBL In use Models answer Korean legal knowledge and reasoning tasks and Korean bar exam questions. It has 7 knowledge tasks, 4 reasoning tasks and 2,510 bar exam items. Solar Pro 4 77.5%
  • KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. Solar Pro 4 93.2%