Benchmark Almanac

LawBench

GPT4 (snapshot unspecified) 53.9 tops its leaderboard GPT-3.5 Turbo Jun 2023 44.5 leads the models we track
Not in model cards

Models answer Chinese legal questions across 20 tasks, including statute recall and case prediction. Each task has its own scoring rule.

GPT-4 led the September 2023 comparison with averages of 52.35 zero-shot and 53.85 one-shot. The authors also tested GPT-3.5 Turbo, StableBeluga2, Qwen, and InternLM.

Made by
Zhiwei Fei and colleagues
Released
September 2023
Kind
Law
Level
University
Scored
Checked automatically
Answers
Short answer
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
GPT-3.5 Turbo Jun 2023 #209 on the capability index
Last model added
September 2023
Scores refreshed
November 20th, 2026
51
Entries on its own leaderboard, 11 models we track

GPT4 (snapshot unspecified) tops the leaderboard with 53.9 GPT-3.5 Turbo leads with 44.5 among models released by September 2023

Best score per entry 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. 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 GPT-3.5 Turbo, at 44.5. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2023, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT4 (snapshot unspecified) one-shot 53.9
  2. 2 GPT-3.5 Turbo Jun 2023, one-shot 44.5
  3. 3 freewilly2_70b-hf zero-shot 39.2
  4. 4 qwen-7b-chat-hf one-shot 39.0
  5. 5 internlm-chat-7b-8k-hf one-shot 37.3
  6. 6 internlm-chat-7b-hf one-shot 36.1
  7. 7 yulan-chat-2-13b-fp16-hf one-shot 34.5
  8. 8 fuzi-mingcha-7b-hf zero-shot 33.0
  9. 9 chatlaw-13b-hf zero-shot 32.8
  10. 10 wisdom-interrogatory-7b-hf zero-shot 31.4
  11. 11 belle-2-13b-hf one-shot 30.8
  12. 12 hanfei-1.0-7b-hf zero-shot 29.7
  13. 13 baichuan-13b-chat-hf one-shot 29.6
  14. 14 lexilaw-6b-hf zero-shot 28.8
  15. 15 chatlaw-33b-hf zero-shot 26.1
  16. 16 tigerbot-sft-7b-hf zero-shot 25.9
  17. 17 lawyer-llama-13b-hf zero-shot 25.3
  18. 18 baichuan-13b-base-hf one-shot 24.0
  19. 19 chatglm2-6b one-shot 22.9
  20. 20 internlm-7b one-shot 22.5
  21. 21 tigerbot-base-7b-hf zero-shot 21.7
  22. 22 baichuan-7b-hf one-shot 21.2
  23. 23 Llama 2-70B one-shot 20.8
  24. 24 Qwen-7B zero-shot 19.6
  25. 25 chinese-alpaca-2-7b-hf one-shot 19.2
  26. 26 vicuna-33b-hf one-shot 19.1
  27. 27 ziya-llama-13b-hf one-shot 18.8
  28. 28 Llama 2-13B one-shot 18.4
  29. 29 Llama 2-7B one-shot 18.3
  30. 30 LLaMA-65B one-shot 17.7
  31. 31 llama-2-chinese-13b-hf one-shot 17.1
  32. 32 LLaMA-13B one-shot 16.6
  33. 33 llama-30b one-shot 16.5
  34. 34 gogpt-7b one-shot 16.2
  35. 35 vicuna-13b-hf one-shot 15.8
  36. 36 moss-moon-003-base-hf one-shot 15.7
  37. 37 alpaca-7b-hf one-shot 15.6
  38. 38 moss-moon-003-sft-hf one-shot 15.6
  39. 39 LLaMA-7B one-shot 14.8
  40. 40 llama-2-70b-chat zero-shot 14.7
  41. 41 llama-2-13b-chat zero-shot 14.6
  42. 42 wizardlm-7b-hf one-shot 14.5
  43. 43 MPT-7B one-shot 14.3
  44. 44 xverse-13b-hf one-shot 13.7
  45. 45 vicuna-7b-hf one-shot 13.7
  46. 46 mpt-instruct-7b-hf one-shot 12.6
  47. 47 chinese-llama-2-7b-hf one-shot 12.6
  48. 48 lawgpt-7b-beta1.1-hf one-shot 11.6
  49. 49 llama-2-chinese-7b-hf one-shot 10.7
  50. 50 llama-2-7b-chat zero-shot 7.16
  51. 51 lawgpt-7b-beta1.0-hf one-shot 2.33

Source: Official LawBench leaderboard

#EntryScoreSettingDateSource
1 GPT4 (snapshot unspecified) 53.9 one-shot Official leaderboard
2 GPT-3.5 Turbo Jun 2023 44.5 one-shot Jun 2023 Official leaderboard
3 freewilly2_70b-hf 39.2 zero-shot Official leaderboard
4 qwen-7b-chat-hf 39.0 one-shot Official leaderboard
5 internlm-chat-7b-8k-hf 37.3 one-shot Official leaderboard
6 internlm-chat-7b-hf 36.1 one-shot Official leaderboard
7 yulan-chat-2-13b-fp16-hf 34.5 one-shot Official leaderboard
8 fuzi-mingcha-7b-hf 33.0 zero-shot Official leaderboard
9 chatlaw-13b-hf 32.8 zero-shot Official leaderboard
10 wisdom-interrogatory-7b-hf 31.4 zero-shot Official leaderboard
11 belle-2-13b-hf 30.8 one-shot Official leaderboard
12 hanfei-1.0-7b-hf 29.7 zero-shot Official leaderboard
13 baichuan-13b-chat-hf 29.6 one-shot Official leaderboard
14 lexilaw-6b-hf 28.8 zero-shot Official leaderboard
15 chatlaw-33b-hf 26.1 zero-shot Official leaderboard
16 tigerbot-sft-7b-hf 25.9 zero-shot Official leaderboard
17 lawyer-llama-13b-hf 25.3 zero-shot Official leaderboard
18 baichuan-13b-base-hf 24.0 one-shot Official leaderboard
19 chatglm2-6b 22.9 one-shot Jun 2023 Official leaderboard
20 internlm-7b 22.5 one-shot Jul 2023 Official leaderboard
21 tigerbot-base-7b-hf 21.7 zero-shot Official leaderboard
22 baichuan-7b-hf 21.2 one-shot Official leaderboard
23 Llama 2-70B 20.8 one-shot Jul 2023 Official leaderboard
24 Qwen-7B 19.6 zero-shot Sep 2023 Official leaderboard
25 chinese-alpaca-2-7b-hf 19.2 one-shot Official leaderboard
26 vicuna-33b-hf 19.1 one-shot Official leaderboard
27 ziya-llama-13b-hf 18.8 one-shot Official leaderboard
28 Llama 2-13B 18.4 one-shot Jul 2023 Official leaderboard
29 Llama 2-7B 18.3 one-shot Jul 2023 Official leaderboard
30 LLaMA-65B 17.7 one-shot Feb 2023 Official leaderboard
31 llama-2-chinese-13b-hf 17.1 one-shot Official leaderboard
32 LLaMA-13B 16.6 one-shot Feb 2023 Official leaderboard
33 llama-30b 16.5 one-shot Official leaderboard
34 gogpt-7b 16.2 one-shot Official leaderboard
35 vicuna-13b-hf 15.8 one-shot Official leaderboard
36 moss-moon-003-base-hf 15.7 one-shot Official leaderboard
37 alpaca-7b-hf 15.6 one-shot Official leaderboard
38 moss-moon-003-sft-hf 15.6 one-shot Official leaderboard
39 LLaMA-7B 14.8 one-shot Feb 2023 Official leaderboard
40 llama-2-70b-chat 14.7 zero-shot Official leaderboard
41 llama-2-13b-chat 14.6 zero-shot Official leaderboard
42 wizardlm-7b-hf 14.5 one-shot Official leaderboard
43 MPT-7B 14.3 one-shot May 2023 Official leaderboard
44 xverse-13b-hf 13.7 one-shot Official leaderboard
45 vicuna-7b-hf 13.7 one-shot Official leaderboard
46 mpt-instruct-7b-hf 12.6 one-shot Official leaderboard
47 chinese-llama-2-7b-hf 12.6 one-shot Official leaderboard
48 lawgpt-7b-beta1.1-hf 11.6 one-shot Official leaderboard
49 llama-2-chinese-7b-hf 10.7 one-shot Official leaderboard
50 llama-2-7b-chat 7.16 zero-shot Official leaderboard
51 lawgpt-7b-beta1.0-hf 2.33 one-shot Official leaderboard
#ModelScoreSettingDateSource
1 GPT-3.5 Turbo Jun 2023 44.5 one-shot Jun 2023 Official leaderboard
2 chatglm2-6b 22.9 one-shot Jun 2023 Official leaderboard
3 internlm-7b 22.5 one-shot Jul 2023 Official leaderboard
4 Llama 2-70B 20.8 one-shot Jul 2023 Official leaderboard
5 Qwen-7B 19.6 zero-shot Sep 2023 Official leaderboard
6 Llama 2-13B 18.4 one-shot Jul 2023 Official leaderboard
7 Llama 2-7B 18.3 one-shot Jul 2023 Official leaderboard
8 LLaMA-65B 17.7 one-shot Feb 2023 Official leaderboard
9 LLaMA-13B 16.6 one-shot Feb 2023 Official leaderboard
10 LLaMA-7B 14.8 one-shot Feb 2023 Official leaderboard
11 MPT-7B 14.3 one-shot May 2023 Official leaderboard

Rankings by part

We read 20 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Task 1-1, Task 1-2 and Task 2-1. GPT-3.5 Turbo leads 16 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Task 1-1

11 models we track have a score here.

  1. 1 GPT-3.5 Turbo Jun 2023 16.1
  2. 2 chatglm2-6b 15.4
  3. 3 Llama 2-13B ↑3 3.58
  4. 4 MPT-7B ↑7 3.57
  5. 5 Llama 2-7B ↑2 3.42
  6. 6 Llama 2-70B ↓2 3.33
  7. 7 LLaMA-7B ↑3 3.14
  8. 8 LLaMA-13B ↑1 3.00
  9. 9 internlm-7b ↓6 2.79
  10. 10 LLaMA-65B ↓2 2.00

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
GPT-3.5 Turbo
zero-shot
44.5
chatglm2-6b
zero-shot
22.9
internlm-7b
zero-shot one-shot
22.5
Llama 2-70B
zero-shot one-shot
20.8
Qwen-7B
one-shot
19.6
Llama 2-13B
zero-shot one-shot
18.4
Llama 2-7B
zero-shot one-shot
18.3
LLaMA-65B
zero-shot one-shot
17.7
LLaMA-13B
zero-shot one-shot
16.6
LLaMA-7B
zero-shot one-shot
14.8
MPT-7B
zero-shot
14.3

Source: Official LawBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-3.5 Turbo Jun 2023 one-shot 44.5
GPT-3.5 Turbo Jun 2023 zero-shot 42.1
chatglm2-6b one-shot 22.9
internlm-7b one-shot 22.5
chatglm2-6b zero-shot 21.1
Llama 2-70B one-shot 20.8
Qwen-7B zero-shot 19.6
Qwen-7B one-shot 18.6
Llama 2-13B one-shot 18.4
Llama 2-7B one-shot 18.3
LLaMA-65B one-shot 17.7
LLaMA-13B one-shot 16.6
internlm-7b zero-shot 16.3
LLaMA-7B one-shot 14.8
MPT-7B one-shot 14.3
Llama 2-70B zero-shot 13.6
Llama 2-7B zero-shot 11.7
MPT-7B zero-shot 11.2
Llama 2-13B zero-shot 9.78
LLaMA-7B zero-shot 9.72
LLaMA-13B zero-shot 8.74
LLaMA-65B zero-shot 8.44

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
GPT-3.5 Turbo
zero-shot
44.5
chatglm2-6b
zero-shot
22.9
internlm-7b
zero-shot one-shot
22.5
Llama 2-70B
zero-shot one-shot
20.8
Qwen-7B
one-shot
19.6
Llama 2-13B
zero-shot one-shot
18.4
Llama 2-7B
zero-shot one-shot
18.3
LLaMA-65B
zero-shot one-shot
17.7
LLaMA-13B
zero-shot one-shot
16.6
LLaMA-7B
zero-shot one-shot
14.8
MPT-7B
zero-shot
14.3

Source: Official LawBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-3.5 Turbo Jun 2023 one-shot 44.5
GPT-3.5 Turbo Jun 2023 zero-shot 42.1
chatglm2-6b one-shot 22.9
internlm-7b one-shot 22.5
chatglm2-6b zero-shot 21.1
Llama 2-70B one-shot 20.8
Qwen-7B zero-shot 19.6
Qwen-7B one-shot 18.6
Llama 2-13B one-shot 18.4
Llama 2-7B one-shot 18.3
LLaMA-65B one-shot 17.7
LLaMA-13B one-shot 16.6
internlm-7b zero-shot 16.3
LLaMA-7B one-shot 14.8
MPT-7B one-shot 14.3
Llama 2-70B zero-shot 13.6
Llama 2-7B zero-shot 11.7
MPT-7B zero-shot 11.2
Llama 2-13B zero-shot 9.78
LLaMA-7B zero-shot 9.72
LLaMA-13B zero-shot 8.74
LLaMA-65B zero-shot 8.44

Number one over time

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

OpenAI
Own leaderboard Not tested since Sep 2023

Source: Official LawBench 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
0.00 10.0 20.0 30.0 40.0 50.0 2023 0.00 10.0 20.0 30.0 40.0 50.0 2023

Source: Official LawBench 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

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 3 years ago, in September 2023, for Qwen-7B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (51 entries).
  • Pass Models scored 11 models we track, plus 40 other entries on its leaderboard.
  • Warning Newest score The newest came 3 years ago, in September 2023, for Qwen-7B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-3.5 Turbo (#209).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 22 rows for models we track and 40 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