Benchmark Almanac

SnorkelLegal

Gemini 4 Argon 46% leads since October 2026
Not in model cards

Agents work 200 tasks inside a simulated mid-market law firm. Each request ties to a matter, a procedural stage and a role with set authority.

15 simulated personas. Record changes are checked by code and legal work product by scenario rubrics. The board reports pass@1.

Made by
Snorkel AI
Released
October 2026
Kind
Law
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
16
Models scored on its own leaderboard

Gemini 4 Argon leads with 46%

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.

16 models. Scroll sideways for the rest.

  1. 46.0 Gemini 4 Argon
  2. 40.7 Grok 4.6
  3. 38.3 Muse Wonder
  4. 37.8 GLM-5.3
  5. 36.5 Muse Spark 1.3
  6. 36.1 Kimi K3
  7. 35.9 Claude Fable 5.1
  8. 32.8 Grok 4.7
  9. 31.4 Claude Opus 5.5
  10. 30.4 DeepSeek-V4-Pro
  11. 29.4 Nemotron 3 Ultra
  12. 26.4 Claude Opus 5
  13. 25.6 Qwen 3.8 Max
  14. 25.3 Gemini 3.8 Flash
  15. 20.9 GPT-6 Astra
  16. 20.7 GPT-6.1 Sol
  1. 1 Gemini 4 Argon 46%
  2. 2 Grok 4.6 40.7%
  3. 3 Muse Wonder 38.3%
  4. 4 GLM-5.3 37.8%
  5. 5 Muse Spark 1.3 36.5%
  6. 6 Kimi K3 36.1%
  7. 7 Claude Fable 5.1 35.9%
  8. 8 Grok 4.7 32.8%
  9. 9 Claude Opus 5.5 31.4%
  10. 10 DeepSeek-V4-Pro 30.4%
  11. 11 Nemotron 3 Ultra 29.4%
  12. 12 Claude Opus 5 26.4%
  13. 13 Qwen 3.8 Max 25.6%
  14. 14 Gemini 3.8 Flash 25.3%
  15. 15 GPT-6 Astra 20.9%
  16. 16 GPT-6.1 Sol 20.7%

Source: Official SnorkelLegal leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Argon 46% Sep 2026 Official leaderboard
2 Grok 4.6 40.7% Aug 2026 Official leaderboard
3 Muse Wonder 38.3% Oct 2026 Official leaderboard
4 GLM-5.3 37.8% Aug 2026 Official leaderboard
5 Muse Spark 1.3 36.5% Sep 2026 Official leaderboard
6 Kimi K3 36.1% Jul 2026 Official leaderboard
7 Claude Fable 5.1 35.9% Sep 2026 Official leaderboard
8 Grok 4.7 32.8% Sep 2026 Official leaderboard
9 Claude Opus 5.5 31.4% Sep 2026 Official leaderboard
10 DeepSeek-V4-Pro 30.4% Apr 2026 Official leaderboard
11 Nemotron 3 Ultra 29.4% Jun 2026 Official leaderboard
12 Claude Opus 5 26.4% Jul 2026 Official leaderboard
13 Qwen 3.8 Max 25.6% Aug 2026 Official leaderboard
14 Gemini 3.8 Flash 25.3% Sep 2026 Official leaderboard
15 GPT-6 Astra 20.9% Sep 2026 Official leaderboard
16 GPT-6.1 Sol 20.7% Sep 2026 Official leaderboard

Pass@1 vs. cost per trial

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

20% 30% 40% 50% $0$2.5$5$7.5$10$12.5 Cost per trial ($), lower is better Pass@1 DeepSeek-V4-Pro: Pass@1 30.4%, cost per trial $1.37 Kimi K3: Pass@1 36.1%, cost per trial $1.94 Claude Opus 5: Pass@1 26.4%, cost per trial $4.14 Qwen 3.8 Max: Pass@1 25.6%, cost per trial $2.34 Claude Fable 5.1: Pass@1 35.9%, cost per trial $5.26 Gemini 3.8 Flash: Pass@1 25.3%, cost per trial $1.23 Muse Spark 1.3: Pass@1 36.5%, cost per trial $1.45 GPT-6 Astra: Pass@1 20.9%, cost per trial $11.6 Grok 4.7: Pass@1 32.8%, cost per trial $3.67 Claude Opus 5.5: Pass@1 31.4%, cost per trial $2.07 GPT-6.1 Sol: Pass@1 20.7%, cost per trial $1.43 Muse Wonder: Pass@1 38.3%, cost per trial $7.24 Nemotron 3 Ultra: Pass@1 29.4%, cost per trial $0.4 Grok 4.6: Pass@1 40.7%, cost per trial $2.22 GLM-5.3: Pass@1 37.8%, cost per trial $1.37 Gemini 4 Argon: Pass@1 46%, cost per trial $3.05 Gemini 4 ArgonGrok 4.6GLM-5.3Nemotron 3 Ultra 20% 30% 40% 50% $0$2.5$5$7.5$10$12.5 Cost per trial ($), lower is better Pass@1 DeepSeek-V4-Pro: Pass@1 30.4%, cost per trial $1.37 Kimi K3: Pass@1 36.1%, cost per trial $1.94 Claude Opus 5: Pass@1 26.4%, cost per trial $4.14 Qwen 3.8 Max: Pass@1 25.6%, cost per trial $2.34 Claude Fable 5.1: Pass@1 35.9%, cost per trial $5.26 Gemini 3.8 Flash: Pass@1 25.3%, cost per trial $1.23 Muse Spark 1.3: Pass@1 36.5%, cost per trial $1.45 GPT-6 Astra: Pass@1 20.9%, cost per trial $11.6 Grok 4.7: Pass@1 32.8%, cost per trial $3.67 Claude Opus 5.5: Pass@1 31.4%, cost per trial $2.07 GPT-6.1 Sol: Pass@1 20.7%, cost per trial $1.43 Muse Wonder: Pass@1 38.3%, cost per trial $7.24 Nemotron 3 Ultra: Pass@1 29.4%, cost per trial $0.4 Grok 4.6: Pass@1 40.7%, cost per trial $2.22 GLM-5.3: Pass@1 37.8%, cost per trial $1.37 Gemini 4 Argon: Pass@1 46%, cost per trial $3.05 Gemini 4 ArgonGrok 4.6GLM-5.3Nemotron 3 Ultra
ModelSettingPass@1Cost per trial
Gemini 4 Argon 46% $3.05
Grok 4.6 40.7% $2.22
Muse Wonder 38.3% $7.24
GLM-5.3 37.8% $1.37
Muse Spark 1.3 36.5% $1.45
Kimi K3 36.1% $1.94
Claude Fable 5.1 35.9% $5.26
Grok 4.7 32.8% $3.67
Claude Opus 5.5 31.4% $2.07
DeepSeek-V4-Pro 30.4% $1.37
Nemotron 3 Ultra 29.4% $0.4
Claude Opus 5 26.4% $4.14
Qwen 3.8 Max 25.6% $2.34
Gemini 3.8 Flash 25.3% $1.23
GPT-6 Astra 20.9% $11.61
GPT-6.1 Sol 20.7% $1.43

The best score, 46%, costs $3.05 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.

NVIDIA xAI Z.ai Google DeepMind Cost frontier One model at each effort level
0% 10% 20% 30% 40% 50% $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Argon Grok 4.6 GLM-5.3 Nemotron 3 Ultra 0% 10% 20% 30% 40% 50% $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 4 Argon Grok 4.6

Source: Official SnorkelLegal leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Argon 46% $3.05
Grok 4.6 40.7% $2.22
Muse Wonder 38.3% $7.24
GLM-5.3 37.8% $1.37
Muse Spark 1.3 36.5% $1.45
Kimi K3 36.1% $1.94
Claude Fable 5.1 35.9% $5.26
Grok 4.7 32.8% $3.67
Claude Opus 5.5 31.4% $2.07
DeepSeek-V4-Pro 30.4% $1.37
Nemotron 3 Ultra 29.4% $0.40
Claude Opus 5 26.4% $4.14
Qwen 3.8 Max 25.6% $2.34
Gemini 3.8 Flash 25.3% $1.23
GPT-6 Astra 20.9% $12
GPT-6.1 Sol 20.7% $1.43

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 46%

Source: Official SnorkelLegal 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 Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Grok 4.6 Kimi K3 DeepSeek-V4-Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Kimi K3

Source: Official SnorkelLegal leaderboard

No lab model card reports it yet

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

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%