Benchmark Almanac

Access to Justice Benchmark

Qwen 3.8 Max 76.4% leads since November 2026
Not in model cards

Models interview a simulated client about a civil legal problem, explain their options, and direct them to local help.

An automated judge checks each response against criteria from legal aid experts across 252 test scenarios.

Made by
Vals AI
Released
November 2026
Kind
Law
Level
Expert
Scored
Graded by a model
Answers
Agent task
Tasks
Kept private
Measures
Capability
Subject
Law
Language
English
In and out
Text → Text
Interaction
Multi-turn
Items from
Generated
Refresh
One fixed set
Run by
A host
In views
Law and finance
Strongest model tested
Claude Opus 5.5 #5 on the capability index
Last model added
October 2026
Scores refreshed
November 20th, 2026
22
Models scored on its own leaderboard

Qwen 3.8 Max leads with 76.4%

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

  1. 1 Qwen 3.8 Max 76.4%
  2. 2 Claude Sonnet 5.5 74.9%
  3. 3 Muse Wonder Full 74.7%
  4. 4 Claude Opus 5.5 73.8%
  5. 5 GLM-5.3 73.7%
  6. 6 Grok 4.7 73.7%
  7. 7 MiniMax-M3.1-Flash 73.3%
  8. 8 DeepSeek V4.1 Flash 1017 73.1%
  9. 9 Muse Wonder 73%
  10. 10 DeepSeek V4.1 Flash 72.9%
  11. 11 Muse Spark 1.3 72.2%
  12. 12 Grok 4.6 71.3%
  13. 13 GLM-5.3-Flash 71.1%
  14. 14 MiMo-V2.6 Flash 70.8%
  15. 15 MiMo-V2.6 Pro 70%
  16. 16 MiniMax-M3 69.8%
  17. 17 Kimi K3.1 Pro 69.5%
  18. 18 Gemini 4 Argon 68.5%
  19. 19 Kimi K3 68.2%
  20. 20 Gemini 3.8 Flash 67.1%
  21. 21 GPT-6.1 Sol 65.8%
  22. 22 GPT-5.6 Luna 65.6%

Source: Official Access to Justice Benchmark leaderboard

#ModelScore ±SettingDateSource
1 Qwen 3.8 Max 76.4% ±1.3 Aug 2026 Official leaderboard
2 Claude Sonnet 5.5 74.9% ±1.4 Sep 2026 Official leaderboard
3 Muse Wonder Full 74.7% ±1.4 Oct 2026 Official leaderboard
4 Claude Opus 5.5 73.8% ±1.4 Sep 2026 Official leaderboard
5 GLM-5.3 73.7% ±1.4 Aug 2026 Official leaderboard
6 Grok 4.7 73.7% ±1.4 Sep 2026 Official leaderboard
7 MiniMax-M3.1-Flash 73.3% ±1.4 Oct 2026 Official leaderboard
8 DeepSeek V4.1 Flash 1017 73.1% ±1.4 Oct 2026 Official leaderboard
9 Muse Wonder 73% ±1.4 Oct 2026 Official leaderboard
10 DeepSeek V4.1 Flash 72.9% ±1.4 Sep 2026 Official leaderboard
11 Muse Spark 1.3 72.2% ±1.4 Sep 2026 Official leaderboard
12 Grok 4.6 71.3% ±1.4 Aug 2026 Official leaderboard
13 GLM-5.3-Flash 71.1% ±1.4 Aug 2026 Official leaderboard
14 MiMo-V2.6 Flash 70.8% ±1.4 Sep 2026 Official leaderboard
15 MiMo-V2.6 Pro 70% ±1.4 Sep 2026 Official leaderboard
16 MiniMax-M3 69.8% ±1.4 Jun 2026 Official leaderboard
17 Kimi K3.1 Pro 69.5% ±1.5 Oct 2026 Official leaderboard
18 Gemini 4 Argon 68.5% ±1.5 Sep 2026 Official leaderboard
19 Kimi K3 68.2% ±1.5 Jul 2026 Official leaderboard
20 Gemini 3.8 Flash 67.1% ±1.5 Sep 2026 Official leaderboard
21 GPT-6.1 Sol 65.8% ±1.5 Sep 2026 Official leaderboard
22 GPT-5.6 Luna 65.6% ±1.5 Jul 2026 Official leaderboard

The best score, 76.4%, costs $0.61 a task

Runs on the cost frontier score higher than any run that's cheaper. MiniMax-M3.1-Flash comes within 3.1 points of the best for $0.06 a task, 10 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Qwen Z.ai MiniMax 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 Qwen 3.8 Max MiniMax-M3.1-Flash GLM-5.3-Flash 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 $10 Cost per task, log scale Qwen 3.8 Max

Source: Official Access to Justice Benchmark leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen 3.8 Max 76.4% $0.61 36,025
Claude Sonnet 5.5 74.9% $2.37 139,888
Muse Wonder Full 74.7% $1.24 26,835
Claude Opus 5.5 73.8% $3.38 93,168
GLM-5.3 73.7% $0.46 41,102
Grok 4.7 73.7% $0.82 24,730
MiniMax-M3.1-Flash 73.3% $0.06 49,568
DeepSeek V4.1 Flash 1017 73.1% $0.09 42,537
Muse Wonder 73% $1.10 22,832
DeepSeek V4.1 Flash 72.9% $0.11 62,223
Muse Spark 1.3 72.2% $0.18 15,949
Muse Spark 1.3 max 72.2% $0.22 17,936
Grok 4.6 71.3% $0.71 15,937
GLM-5.3-Flash 71.1% $0.02 25,469
MiMo-V2.6 Flash 70.8% $0.02 26,645
MiMo-V2.6 Pro 70% $0.05 25,425
MiniMax-M3 69.8% $0.11 15,409
Kimi K3.1 Pro 69.5% $0.96 35,732
Gemini 4 Argon 68.5% $0.83 21,195
Kimi K3 68.2% $0.45 19,747
Gemini 3.8 Flash 67.1% $0.32 17,820
GPT-6.1 Sol 65.8% $0.69 31,157
GPT-5.6 Luna 65.6% $0.09 27,251

Number one over time

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

Qwen
Own leaderboard 1 leader
Qwen 3.8 Max 76.4%

Source: Official Access to Justice Benchmark leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Qwen 3.8 Max MiniMax-M3 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Qwen 3.8 Max MiniMax-M3

Source: Official Access to Justice Benchmark leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Law, finance and medicine

All 637 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%
  • 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. o1-preview 92.7%
  • Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%

Data audit

All 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (22 entries).
  • Pass Models scored 22 models we track.
  • Pass Newest score The newest is from October 2026, for Kimi K3.1 Pro.
  • Pass Frontier models 4 of the top 10 have a score, led by Claude Opus 5.5 (#5).
  • Pass Headroom The best is 76.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on vals.ai: 23 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check