Benchmark Almanac

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

Gemini 4 Argon 27.1% leads since Oct 2026 Not in model cards

Agents produce legal work product from matter files with file tools and docx, pptx and xlsx skills. A task passes only if every expert rubric criterion passes.

Harvey fixed the 120-task private holdout, removing 17 unsolvable criteria. The harness was resynced with a finish tool and a reasoning-first judge. Vals says scores are not comparable with the v1 board.

Made by Harvey (board run by Vals AI) Released Oct 2026 Law Public tasks plus a private test set Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
harveyai/harvey-labs 1,546 stars Paper Leaderboard
33
Models scored on its own leaderboard

Gemini 4 Argon leads with 27.1%

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

33 models. Scroll sideways for the rest.

  1. 27.1 Gemini 4 Argon
  2. 25.0 Muse Spark Open
  3. 23.8 Muse Wonder
  4. 23.3 Muse Wonder Full
  5. 18.3 Grok 4.7
  6. 17.5 Mistral Large 4
  7. 17.5 Gemini 4 Carbon
  8. 15.8 Mistral Large 4 (Le Chonk, open weights)
  9. 15.4 Qwen3.8 Max (0902)
  10. 14.6 GLM-5.3
  11. 14.2 DeepSeek V4.1 Flash 1017
  12. 13.8 MiMo-V2.6 Pro
  13. 13.3 MiMo-V2.6 Flash
  14. 12.9 Kimi K3
  15. 12.9 Step 5 Preview
  16. 12.1 Gemini 3.8 Flash
  17. 11.7 Kimi K3.1 Pro
  18. 10.8 DeepSeek V4 Pro 0813
  19. 10.4 GLM-5.3-Flash
  20. 10.4 Claude Sonnet 5.5
  21. 9.17 Claude Opus 5.5
  22. 8.75 Tencent Hy4 preview
  23. 7.50 MiniMax-M3.1-Flash
  24. 7.08 GPT-6 Astra
  25. 6.25 GPT-6.1 Sol
  26. 5.00 Mercury 3
  27. 4.58 Beam 1.0
  28. 4.17 GPT-6 Luna
  29. 4.17 Claude Haiku 5.5
  30. 3.75 Inkling
  31. 3.75 Nemotron 3.5 Super VL
  32. 2.92 Inkling-Small
  33. 1.67 GPT-6.1 Terra
  1. 1 Gemini 4 Argon 27.1%
  2. 2 Muse Spark Open 25%
  3. 3 Muse Wonder 23.8%
  4. 4 Muse Wonder Full 23.3%
  5. 5 Grok 4.7 18.3%
  6. 6 Mistral Large 4 17.5%
  7. 7 Gemini 4 Carbon 17.5%
  8. 8 Mistral Large 4 (Le Chonk, open weights) 15.8%
  9. 9 Qwen3.8 Max (0902) 15.4%
  10. 10 GLM-5.3 14.6%
  11. 11 DeepSeek V4.1 Flash 1017 14.2%
  12. 12 MiMo-V2.6 Pro 13.8%
  13. 13 MiMo-V2.6 Flash 13.3%
  14. 14 Kimi K3 12.9%
  15. 15 Step 5 Preview 12.9%
  16. 16 Gemini 3.8 Flash 12.1%
  17. 17 Kimi K3.1 Pro 11.7%
  18. 18 DeepSeek V4 Pro 0813 10.8%
  19. 19 GLM-5.3-Flash 10.4%
  20. 20 Claude Sonnet 5.5 10.4%
  21. 21 Claude Opus 5.5 9.17%
  22. 22 Tencent Hy4 preview 8.75%
  23. 23 MiniMax-M3.1-Flash 7.50%
  24. 24 GPT-6 Astra 7.08%
  25. 25 GPT-6.1 Sol 6.25%
  26. 26 Mercury 3 5.00%
  27. 27 Beam 1.0 4.58%
  28. 28 GPT-6 Luna 4.17%
  29. 29 Claude Haiku 5.5 4.17%
  30. 30 Inkling 3.75%
  31. 31 Nemotron 3.5 Super VL 3.75%
  32. 32 Inkling-Small 2.92%
  33. 33 GPT-6.1 Terra 1.67%
#ModelScoreSettingDateSource
1 Gemini 4 Argon 27.1% Sep 2026 Official leaderboard
2 Muse Spark Open 25% Oct 2026 Official leaderboard
3 Muse Wonder 23.8% Oct 2026 Official leaderboard
4 Muse Wonder Full 23.3% Oct 2026 Official leaderboard
5 Grok 4.7 18.3% Sep 2026 Official leaderboard
6 Mistral Large 4 17.5% Oct 2026 Official leaderboard
7 Gemini 4 Carbon 17.5% Oct 2026 Official leaderboard
8 Mistral Large 4 (Le Chonk, open weights) 15.8% Oct 2026 Official leaderboard
9 Qwen3.8 Max (0902) 15.4% Sep 2026 Official leaderboard
10 GLM-5.3 14.6% Aug 2026 Official leaderboard
11 DeepSeek V4.1 Flash 1017 14.2% Oct 2026 Official leaderboard
12 MiMo-V2.6 Pro 13.8% Sep 2026 Official leaderboard
13 MiMo-V2.6 Flash 13.3% Sep 2026 Official leaderboard
14 Kimi K3 12.9% Jul 2026 Official leaderboard
15 Step 5 Preview 12.9% Sep 2026 Official leaderboard
16 Gemini 3.8 Flash 12.1% Sep 2026 Official leaderboard
17 Kimi K3.1 Pro 11.7% Oct 2026 Official leaderboard
18 DeepSeek V4 Pro 0813 10.8% Aug 2026 Official leaderboard
19 GLM-5.3-Flash 10.4% Aug 2026 Official leaderboard
20 Claude Sonnet 5.5 10.4% Sep 2026 Official leaderboard
21 Claude Opus 5.5 9.17% Sep 2026 Official leaderboard
22 Tencent Hy4 preview 8.75% Aug 2026 Official leaderboard
23 MiniMax-M3.1-Flash 7.50% Oct 2026 Official leaderboard
24 GPT-6 Astra 7.08% Sep 2026 Official leaderboard
25 GPT-6.1 Sol 6.25% Sep 2026 Official leaderboard
26 Mercury 3 5.00% Oct 2026 Official leaderboard
27 Beam 1.0 4.58% Oct 2026 Official leaderboard
28 GPT-6 Luna 4.17% Sep 2026 Official leaderboard
29 Claude Haiku 5.5 4.17% Oct 2026 Official leaderboard
30 Inkling 3.75% Jul 2026 Official leaderboard
31 Nemotron 3.5 Super VL 3.75% Oct 2026 Official leaderboard
32 Inkling-Small 2.92% Jul 2026 Official leaderboard
33 GPT-6.1 Terra 1.67% Oct 2026 Official leaderboard

The best score, 27.1%, costs $11 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 Google DeepMind Mistral AI Meta DeepSeek 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 Gemini 4 Argon Muse Wonder Mistral Large 4 Mistral Large 4 (Le Chonk, open weights) DeepSeek V4.1 Flash 1017 MiMo-V2.6 Flash 0% 10% 20% 30% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Argon Muse Wonder
ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Argon 27.1% $11
Muse Spark Open 25% $14
Muse Wonder 23.8% $7.08
Muse Wonder Full 23.3% $5.41
Grok 4.7 18.3% $11
Mistral Large 4 17.5% $2.55
Gemini 4 Carbon 17.5% $6.52
Mistral Large 4 (Le Chonk, open weights) 15.8% $0.73
Qwen3.8 Max (0902) 15.4% $3.36
GLM-5.3 14.6% $3.99
DeepSeek V4.1 Flash 1017 14.2% $0.61
MiMo-V2.6 Pro 13.8% $0.22
MiMo-V2.6 Flash 13.3% $0.09
Kimi K3 12.9% $4.14
Step 5 Preview 12.9% $1.13
Gemini 3.8 Flash 12.1% $3.66
Kimi K3.1 Pro 11.7% $9.00
DeepSeek V4 Pro 0813 10.8% $0.80
GLM-5.3-Flash 10.4% $0.32
Claude Sonnet 5.5 10.4% $17
Claude Opus 5.5 9.17% $21
Tencent Hy4 preview 8.75% $0.98
MiniMax-M3.1-Flash 7.50% $0.47
GPT-6 Astra 7.08% $25
GPT-6.1 Sol 6.25% $3.54
Mercury 3 5.00% $0.19
Beam 1.0 4.58%
GPT-6 Luna 4.17% $0.30
Claude Haiku 5.5 4.17% $4.26
Inkling 3.75% $1.25
Nemotron 3.5 Super VL 3.75% $0.37
Inkling-Small 2.92% $0.18
GPT-6.1 Terra 1.67% $1.48

Number one over time

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

Google DeepMind
Own leaderboard 1 leader
Gemini 4 Argon 27.1%

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% 2027 Near the ceiling (85%) Gemini 4 Argon Grok 4.7 GLM-5.3 Inkling 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Gemini 4 Argon Grok 4.7

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 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 Meta Moonshot AI Google DeepMind
Harvey LAB v1.1 (HLAB-Vals-1.1) 6 leaders
Gemini 4 Argon 27.1%
Harvey LAB (Legal Agent Benchmark) Passed 85% in Jul 2026, by Kimi K3
Kimi K3 94.6%
Harvey LAB v1.1 (HLAB-Vals-1.1) 1 leader
Gemini 4 Argon 27.1%
Replaces Harvey LAB (Legal Agent Benchmark)

More in Law, finance and medicine

All 140 in Law, finance and medicine
  • 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%
  • 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 In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
  • 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%
  • PRBench In use Model answers 1,100 expert-written law and finance questions. LLM judge grades against weighted expert rubrics of 10-30 criteria each. Muse Spark 1.3 61.6%