Benchmark Almanac

LexGenius

DeepSeek-R1 (snapshot unspecified) 64.6% tops its leaderboard Qwen2.5-7B 59.7% leads the models we track
Not in model cards

Models answer Chinese legal questions drawn from cases and exams. The authors compare results across areas including legal reasoning, ethics, and judicial practice.

Made by
Wenjin Liu and colleagues
Released
November 2025
Kind
Law
Level
University
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Subject
Law
Language
Chinese
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Law and finance
Strongest model tested
Qwen3-8B #154 on the capability index
Last model added
April 2025
Scores refreshed
November 20th, 2026
12
Entries in its paper, 8 models we track

DeepSeek-R1 (snapshot unspecified) tops the leaderboard with 64.6% Qwen2.5-7B leads with 59.7% among models released by April 2025

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Qwen2.5-7B, at 59.7%. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 DeepSeek-R1 (snapshot unspecified) Chain of thought 64.6%
  2. 2 DeepSeek-V3 (snapshot unspecified) Chain of thought 64.4%
  3. 3 Qwen2.5-7B Naive prompt 59.7%
  4. 4 Qwen3-8B Naive prompt 55.9%
  5. 5 Qwen3-4B Naive prompt 53.6%
  6. 6 GLM-4-9B Chain of thought 53.5%
  7. 7 GPT-4.1 nano Chain of thought 53.4%
  8. 8 GPT-4o mini Chain of thought 52.6%
  9. 9 Qwen2.5-1.5B Chain of thought 48.2%
  10. 10 LLaMA-3.2-8B Chain of thought 43.1%
  11. 11 DeepSeek-7B Naive prompt 38.6%
  12. 12 Llama 3.2 1B Naive prompt 35.3%

Source: The LexGenius paper

#EntryScoreSettingDateSource
1 DeepSeek-R1 (snapshot unspecified) 64.6% Chain of thought Official leaderboard
2 DeepSeek-V3 (snapshot unspecified) 64.4% Chain of thought Official leaderboard
3 Qwen2.5-7B 59.7% Naive prompt Sep 2024 Official leaderboard
4 Qwen3-8B 55.9% Naive prompt Apr 2025 Official leaderboard
5 Qwen3-4B 53.6% Naive prompt Apr 2025 Official leaderboard
6 GLM-4-9B 53.5% Chain of thought Jun 2024 Official leaderboard
7 GPT-4.1 nano 53.4% Chain of thought Apr 2025 Official leaderboard
8 GPT-4o mini 52.6% Chain of thought Jul 2024 Official leaderboard
9 Qwen2.5-1.5B 48.2% Chain of thought Sep 2024 Official leaderboard
10 LLaMA-3.2-8B 43.1% Chain of thought Official leaderboard
11 DeepSeek-7B 38.6% Naive prompt Official leaderboard
12 Llama 3.2 1B 35.3% Naive prompt Sep 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Qwen2.5-7B 59.7% Naive prompt Sep 2024 Official leaderboard
2 Qwen3-8B 55.9% Naive prompt Apr 2025 Official leaderboard
3 Qwen3-4B 53.6% Naive prompt Apr 2025 Official leaderboard
4 GLM-4-9B 53.5% Chain of thought Jun 2024 Official leaderboard
5 GPT-4.1 nano 53.4% Chain of thought Apr 2025 Official leaderboard
6 GPT-4o mini 52.6% Chain of thought Jul 2024 Official leaderboard
7 Qwen2.5-1.5B 48.2% Chain of thought Sep 2024 Official leaderboard
8 Llama 3.2 1B 35.3% Naive prompt Sep 2024 Official leaderboard

Rankings by part

We read 7 parts from the benchmark’s paper, each ranking models on one part of the test, such as Legal Understanding, Legal Reasoning and Legal Application. Qwen2.5-7B leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.

Legal Understanding

8 models we track have a score here.

  1. 1 Qwen2.5-7B 61.4%
  2. 2 Qwen3-8B 58.8%
  3. 3 GPT-4.1 nano ↑2 54.6%
  4. 4 Qwen3-4B ↓1 53.2%
  5. 5 GLM-4-9B ↓1 53%
  6. 6 GPT-4o mini 49.8%
  7. 7 Qwen2.5-1.5B 44%
  8. 8 Llama 3.2 1B 29.4%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Qwen2.5-7B
Chain of th…
59.7%
Qwen3-8B
Chain of th…
55.9%
Qwen3-4B
Chain of th…
53.6%
GLM-4-9B
Naive prompt
53.5%
GPT-4.1 nano
Naive prompt Chain of th…
53.4%
GPT-4o mini
Naive prompt
52.6%
Qwen2.5-1.5B
Naive prompt
48.2%
Llama 3.2 1B
Chain of th…
35.3%

Source: The LexGenius paper

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen2.5-7B Naive prompt 59.7%
Qwen2.5-7B Chain of thought 59.4%
Qwen3-8B Naive prompt 55.9%
Qwen3-8B Chain of thought 55.3%
Qwen3-4B Naive prompt 53.6%
GLM-4-9B Chain of thought 53.5%
GPT-4.1 nano Chain of thought 53.4%
GLM-4-9B Naive prompt 53.2%
Qwen3-4B Chain of thought 53.2%
GPT-4o mini Chain of thought 52.6%
GPT-4o mini Naive prompt 52.5%
Qwen2.5-1.5B Chain of thought 48.2%
GPT-4.1 nano Naive prompt 48.2%
Qwen2.5-1.5B Naive prompt 47.6%
Llama 3.2 1B Naive prompt 35.3%
Llama 3.2 1B Chain of thought 34.9%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Qwen2.5-7B
Chain of th…
59.7%
Qwen3-8B
Chain of th…
55.9%
Qwen3-4B
Chain of th…
53.6%
GLM-4-9B
Naive prompt
53.5%
GPT-4.1 nano
Naive prompt Chain of th…
53.4%
GPT-4o mini
Naive prompt
52.6%
Qwen2.5-1.5B
Naive prompt
48.2%
Llama 3.2 1B
Chain of th…
35.3%

Source: The LexGenius paper

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen2.5-7B Naive prompt 59.7%
Qwen2.5-7B Chain of thought 59.4%
Qwen3-8B Naive prompt 55.9%
Qwen3-8B Chain of thought 55.3%
Qwen3-4B Naive prompt 53.6%
GLM-4-9B Chain of thought 53.5%
GPT-4.1 nano Chain of thought 53.4%
GLM-4-9B Naive prompt 53.2%
Qwen3-4B Chain of thought 53.2%
GPT-4o mini Chain of thought 52.6%
GPT-4o mini Naive prompt 52.5%
Qwen2.5-1.5B Chain of thought 48.2%
GPT-4.1 nano Naive prompt 48.2%
Qwen2.5-1.5B Naive prompt 47.6%
Llama 3.2 1B Naive prompt 35.3%
Llama 3.2 1B Chain of thought 34.9%

Number one over time

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

Qwen
Benchmark paper Not tested since Nov 2025
Qwen2.5-7B 59.7%

Source: The LexGenius paper

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
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Qwen2.5-7B GLM-4-9B 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Qwen2.5-7B

Source: Official LexGenius 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

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 19 months ago, in April 2025, for Qwen3-4B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (12 entries).
  • Pass Models scored 8 models we track, plus 4 other entries on its leaderboard.
  • Warning Newest score The newest came 19 months ago, in April 2025, for Qwen3-4B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Qwen3-8B (#154).
  • Pass Headroom The best is 59.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 16 rows for models we track and 4 other entries 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