VLegal-Bench
Models answer questions and solve legal scenarios under Vietnamese law. Legal experts check that reference answers follow authoritative documents.
- Made by
- Nguyen Tien Dong and colleagues
- Released
- December 2025
- Kind
- Law
- Level
- University
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Subject
- Law
- Language
- Vietnamese
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Law and finance
- Strongest model tested
- Claude Sonnet 4.5 #81 on the capability index
- Last model added
- September 2025
- Scores refreshed
- November 20th, 2026
Gemini 2.5 Flash leads with 72% among models released by September 2025 Gemini 2.5 Flash leads with 72% among models released by September 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper gives no overall score. The headline is legal entity recognition accuracy in percent, the first result in Table 3. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in September 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 72.0
Gemini 2.5 Flash
- 71.3 Qwen2.5-32B-Instruct
- 70.0 gpt-4o (snapshot unspecified)
- 69.8 Claude Sonnet 4.5
- 68.6 Qwen2.5-14B-Instruct
- 68.0
Qwen2.5-72B
- 66.2 CMC-Legal-32B
- 65.2 GPT-4o mini
- 62.8 Qwen3-4b-legal-pretrain
- 62.2 SeaLLMs-v3-7B-Chat
- 57.9 gemma-2-27b-it
- 57.2
InternLM3-8B-Instruct
- 55.9 Llama-3.1-8B-Instruct
- 55.7 Llama-3.1-70B-Instruct
- 52.9 gemma-2-9b-it
- 52.8 Qwen2.5-3B-Instruct
- 50.4 Qwen2.5-7B-Instruct
- 47.7 SeaLLMs-v3-1.5B-Chat
- 46.7 BloomVN-8B-chat
- 25.5 Llama-2-7b-chat-hf
- 22.7 Llama-2-13b-chat-hf
- 22.7 gpt-oss-20b
- 16.4 internlm-chat-20b
- 1
Gemini 2.5 Flash Sep 2025, Zero-shot without reasoning 72%
- 2 Qwen2.5-32B-Instruct Zero-shot without reasoning 71.3%
- 3 gpt-4o (snapshot unspecified) Zero-shot without reasoning 70%
- 4 Claude Sonnet 4.5 Zero-shot without reasoning 69.8%
- 5 Qwen2.5-14B-Instruct Zero-shot without reasoning 68.6%
- 6
Qwen2.5-72B Zero-shot without reasoning 68%
- 7 CMC-Legal-32B Zero-shot without reasoning 66.2%
- 8 GPT-4o mini Zero-shot without reasoning 65.2%
- 9 Qwen3-4b-legal-pretrain Zero-shot without reasoning 62.8%
- 10 SeaLLMs-v3-7B-Chat Zero-shot without reasoning 62.2%
- 11 gemma-2-27b-it Zero-shot without reasoning 57.9%
- 12
InternLM3-8B-Instruct Zero-shot without reasoning 57.2%
- 13 Llama-3.1-8B-Instruct Zero-shot without reasoning 55.9%
- 14 Llama-3.1-70B-Instruct Zero-shot without reasoning 55.7%
- 15 gemma-2-9b-it Zero-shot without reasoning 52.9%
- 16 Qwen2.5-3B-Instruct Zero-shot without reasoning 52.8%
- 17 Qwen2.5-7B-Instruct Zero-shot without reasoning 50.4%
- 18 SeaLLMs-v3-1.5B-Chat Zero-shot without reasoning 47.7%
- 19 BloomVN-8B-chat Zero-shot without reasoning 46.7%
- 20 Llama-2-7b-chat-hf Zero-shot without reasoning 25.5%
- 21 Llama-2-13b-chat-hf Zero-shot without reasoning 22.7%
- 22 gpt-oss-20b Zero-shot without reasoning 22.7%
- 23 internlm-chat-20b Zero-shot without reasoning 16.4%
Source: The VLegal-Bench paper
- 1
Gemini 2.5 Flash Sep 2025, Zero-shot without reasoning 72%
- 2 Claude Sonnet 4.5 Zero-shot without reasoning 69.8%
- 3
Qwen2.5-72B Zero-shot without reasoning 68%
- 4 GPT-4o mini Zero-shot without reasoning 65.2%
- 5
InternLM3-8B-Instruct Zero-shot without reasoning 57.2%
- 6 gpt-oss-20b Zero-shot without reasoning 22.7%
- 7 internlm-chat-20b Zero-shot without reasoning 16.4%
Source: The VLegal-Bench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 72% | Zero-shot without reasoning | Sep 2025 | Official leaderboard |
| 2 | Qwen2.5-32B-Instruct | 71.3% | Zero-shot without reasoning | Official leaderboard | |
| 3 | gpt-4o (snapshot unspecified) | 70% | Zero-shot without reasoning | Official leaderboard | |
| 4 | Claude Sonnet 4.5 | 69.8% | Zero-shot without reasoning | Sep 2025 | Official leaderboard |
| 5 | Qwen2.5-14B-Instruct | 68.6% | Zero-shot without reasoning | Official leaderboard | |
| 6 | | 68% | Zero-shot without reasoning | Sep 2024 | Official leaderboard |
| 7 | CMC-Legal-32B | 66.2% | Zero-shot without reasoning | Official leaderboard | |
| 8 | GPT-4o mini | 65.2% | Zero-shot without reasoning | Jul 2024 | Official leaderboard |
| 9 | Qwen3-4b-legal-pretrain | 62.8% | Zero-shot without reasoning | Official leaderboard | |
| 10 | SeaLLMs-v3-7B-Chat | 62.2% | Zero-shot without reasoning | Official leaderboard | |
| 11 | gemma-2-27b-it | 57.9% | Zero-shot without reasoning | Official leaderboard | |
| 12 | | 57.2% | Zero-shot without reasoning | Jan 2025 | Official leaderboard |
| 13 | Llama-3.1-8B-Instruct | 55.9% | Zero-shot without reasoning | Official leaderboard | |
| 14 | Llama-3.1-70B-Instruct | 55.7% | Zero-shot without reasoning | Official leaderboard | |
| 15 | gemma-2-9b-it | 52.9% | Zero-shot without reasoning | Official leaderboard | |
| 16 | Qwen2.5-3B-Instruct | 52.8% | Zero-shot without reasoning | Official leaderboard | |
| 17 | Qwen2.5-7B-Instruct | 50.4% | Zero-shot without reasoning | Official leaderboard | |
| 18 | SeaLLMs-v3-1.5B-Chat | 47.7% | Zero-shot without reasoning | Official leaderboard | |
| 19 | BloomVN-8B-chat | 46.7% | Zero-shot without reasoning | Official leaderboard | |
| 20 | Llama-2-7b-chat-hf | 25.5% | Zero-shot without reasoning | Official leaderboard | |
| 21 | Llama-2-13b-chat-hf | 22.7% | Zero-shot without reasoning | Official leaderboard | |
| 22 | gpt-oss-20b | 22.7% | Zero-shot without reasoning | Aug 2025 | Official leaderboard |
| 23 | internlm-chat-20b | 16.4% | Zero-shot without reasoning | Sep 2023 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 72% | Zero-shot without reasoning | Sep 2025 | Official leaderboard |
| 2 | Claude Sonnet 4.5 | 69.8% | Zero-shot without reasoning | Sep 2025 | Official leaderboard |
| 3 | | 68% | Zero-shot without reasoning | Sep 2024 | Official leaderboard |
| 4 | GPT-4o mini | 65.2% | Zero-shot without reasoning | Jul 2024 | Official leaderboard |
| 5 | | 57.2% | Zero-shot without reasoning | Jan 2025 | Official leaderboard |
| 6 | gpt-oss-20b | 22.7% | Zero-shot without reasoning | Aug 2025 | Official leaderboard |
| 7 | internlm-chat-20b | 16.4% | Zero-shot without reasoning | Sep 2023 | Official leaderboard |
Rankings by part
We read 23 parts from the benchmark’s paper, each ranking models on one part of the test, such as Legal Entity Recognition (Acc), Legal Topic Classification (Acc) and Legal Concept Recall (Acc). Claude Sonnet 4.5 leads 12 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Legal Entity Recognition (Acc)
7 models we track have a score here.
- 1
Gemini 2.5 Flash Sep 2025 72%
- 2 Claude Sonnet 4.5 69.8%
- 3
Qwen2.5-72B 68%
- 4 GPT-4o mini 65.2%
- 5
InternLM3-8B-Instruct 57.2%
- 6 gpt-oss-20b 22.7%
- 7 internlm-chat-20b 16.4%
Legal Topic Classification (Acc)
7 models we track have a score here.
- 1 GPT-4o mini ↑3 82.7%
- 2 Claude Sonnet 4.5 82.5%
- 3
Gemini 2.5 Flash Sep 2025 ↓2 81.4%
- 4
Qwen2.5-72B ↓1 80.4%
- 5 gpt-oss-20b ↑1 73.2%
- 6
InternLM3-8B-Instruct ↓1 46.3%
- 7 internlm-chat-20b 24%
Legal Concept Recall (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 83%
- 2
Qwen2.5-72B ↑1 77%
- 3 GPT-4o mini ↑1 61.3%
- 4
Gemini 2.5 Flash Sep 2025 ↓3 61.3%
- 5
InternLM3-8B-Instruct 51%
- 6 gpt-oss-20b 47.3%
- 7 internlm-chat-20b 17.9%
Article Recall (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 84.2%
- 2
Gemini 2.5 Flash Sep 2025 ↓1 81.4%
- 3
Qwen2.5-72B 79.1%
- 4 GPT-4o mini 68.4%
- 5
InternLM3-8B-Instruct 55.5%
- 6 gpt-oss-20b 39%
- 7 internlm-chat-20b 21.7%
Legal Schema Recall (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 27.3%
- 2
InternLM3-8B-Instruct ↑3 24.5%
- 3
Gemini 2.5 Flash Sep 2025 ↓2 24.3%
- 4 GPT-4o mini 22.3%
- 5
Qwen2.5-72B ↓2 21.9%
- 6 gpt-oss-20b 18.6%
- 7 internlm-chat-20b 18.1%
Relation Extraction (Acc)
7 models we track have a score here.
- 1 GPT-4o mini ↑3 87.3%
- 2
Qwen2.5-72B ↑1 80.8%
- 3
Gemini 2.5 Flash Sep 2025 ↓2 80.6%
- 4 Claude Sonnet 4.5 ↓2 79.2%
- 5
InternLM3-8B-Instruct 68.4%
- 6 gpt-oss-20b 45.1%
- 7 internlm-chat-20b 11.1%
Legal Element Recognition (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 75.3%
- 2
Qwen2.5-72B ↑1 65.7%
- 3
Gemini 2.5 Flash Sep 2025 ↓2 64%
- 4 GPT-4o mini 51.3%
- 5
InternLM3-8B-Instruct 42.7%
- 6 gpt-oss-20b 28.3%
- 7 internlm-chat-20b 9.00%
Legal Graph Structuring (R-L)
7 models we track have a score here.
- 1
Qwen2.5-72B ↑2 0.81
- 2 Claude Sonnet 4.5 0.81
- 3
Gemini 2.5 Flash Sep 2025 ↓2 0.66
- 4 GPT-4o mini 0.53
- 5
InternLM3-8B-Instruct 0.40
- 6 gpt-oss-20b 0.25
- 7 internlm-chat-20b 0.19
Judgment Verification (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 87.8%
- 2
Qwen2.5-72B ↑1 83.8%
- 3
Gemini 2.5 Flash Sep 2025 ↓2 82.3%
- 4 GPT-4o mini 73.3%
- 5
InternLM3-8B-Instruct 67.3%
- 6 gpt-oss-20b 66.5%
- 7 internlm-chat-20b 61.2%
User Intent Understanding (m-F1)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 62%
- 2 GPT-4o mini ↑2 61.7%
- 3 gpt-oss-20b ↑3 56.4%
- 4
InternLM3-8B-Instruct ↑1 52.3%
- 5
Gemini 2.5 Flash Sep 2025 ↓4 49.9%
- 6 internlm-chat-20b ↑1 32.7%
- 7
Qwen2.5-72B ↓4 23.6%
Article / Clause Prediction (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 59.8%
- 2
Gemini 2.5 Flash Sep 2025 ↓1 40.8%
- 3 GPT-4o mini ↑1 35.3%
- 4
Qwen2.5-72B ↓1 33.5%
- 5 gpt-oss-20b ↑1 29.7%
- 6
InternLM3-8B-Instruct ↓1 26.7%
- 7 internlm-chat-20b 19.7%
Legal Court Decision Prediction (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 88.8%
- 2
Qwen2.5-72B ↑1 85.5%
- 3
Gemini 2.5 Flash Sep 2025 ↓2 84.7%
- 4 GPT-4o mini 82.2%
- 5
InternLM3-8B-Instruct 71%
- 6 gpt-oss-20b 66%
- 7 internlm-chat-20b 37.2%
Multi-Article Reasoning (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 79.1%
- 2
Gemini 2.5 Flash Sep 2025 ↓1 76.7%
- 3 GPT-4o mini ↑1 74.7%
- 4
Qwen2.5-72B ↓1 74.3%
- 5
InternLM3-8B-Instruct 65.1%
- 6 gpt-oss-20b 52.7%
- 7 internlm-chat-20b 23.3%
Conflict & Consistency Detection (Y-F1)
7 models we track have a score here.
- 1 gpt-oss-20b ↑5 73.4%
- 2 internlm-chat-20b ↑5 64.9%
- 3
InternLM3-8B-Instruct ↑2 41%
- 4 GPT-4o mini 11.8%
- 5
Qwen2.5-72B ↓2 0%
- 6
Gemini 2.5 Flash Sep 2025 ↓5 0%
- 7 Claude Sonnet 4.5 ↓5 0%
Conflict & Consistency Detection (N-F1)
7 models we track have a score here.
- 1
Qwen2.5-72B ↑2 39.6%
- 2
Gemini 2.5 Flash Sep 2025 ↓1 39.6%
- 3 Claude Sonnet 4.5 ↓1 39.6%
- 4 GPT-4o mini 39.6%
- 5 gpt-oss-20b ↑1 35.6%
- 6
InternLM3-8B-Instruct ↓1 8.33%
- 7 internlm-chat-20b 0%
Penalty / Remedy Estimation (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 69.4%
- 2
Qwen2.5-72B ↑1 63.5%
- 3
Gemini 2.5 Flash Sep 2025 ↓2 63.5%
- 4 GPT-4o mini 57.4%
- 5
InternLM3-8B-Instruct 43%
- 6 gpt-oss-20b 37.1%
- 7 internlm-chat-20b 32.6%
Legal Document Summarization (R-L)
7 models we track have a score here.
- 1 GPT-4o mini ↑3 0.33
- 2
Qwen2.5-72B ↑1 0.29
- 3 Claude Sonnet 4.5 ↓1 0.28
- 4
Gemini 2.5 Flash Sep 2025 ↓3 0.28
- 5
InternLM3-8B-Instruct 0.26
- 6 internlm-chat-20b ↑1 0.06
- 7 gpt-oss-20b ↓1 0.03
Judicial Reasoning Generation (R-L)
7 models we track have a score here.
- 1 GPT-4o mini ↑3 0.42
- 2
Qwen2.5-72B ↑1 0.41
- 3
Gemini 2.5 Flash Sep 2025 ↓2 0.40
- 4 Claude Sonnet 4.5 ↓2 0.39
- 5
InternLM3-8B-Instruct 0.27
- 6 internlm-chat-20b ↑1 0.26
- 7 gpt-oss-20b ↓1 0.15
Objective Legal Opinion Generation (R-L)
7 models we track have a score here.
- 1 GPT-4o mini ↑3 0.49
- 2
Qwen2.5-72B ↑1 0.48
- 3 Claude Sonnet 4.5 ↓1 0.38
- 4
Gemini 2.5 Flash Sep 2025 ↓3 0.38
- 5 gpt-oss-20b ↑1 0.31
- 6 internlm-chat-20b ↑1 0.29
- 7
InternLM3-8B-Instruct ↓2 0.29
Bias Detection (Acc)
7 models we track have a score here.
- 1
Qwen2.5-72B ↑2 46.6%
- 2
Gemini 2.5 Flash Sep 2025 ↓1 46.2%
- 3 Claude Sonnet 4.5 ↓1 45.4%
- 4 GPT-4o mini 41.8%
- 5 internlm-chat-20b ↑2 29.3%
- 6 gpt-oss-20b 21.7%
- 7
InternLM3-8B-Instruct ↓2 15.6%
Privacy & Data Protection (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 72.8%
- 2
Gemini 2.5 Flash Sep 2025 ↓1 68.7%
- 3 GPT-4o mini ↑1 67.7%
- 4
Qwen2.5-72B ↓1 67.3%
- 5
InternLM3-8B-Instruct 58.8%
- 6 gpt-oss-20b 58.8%
- 7 internlm-chat-20b 24%
Ethical Consistency Assessment (Acc)
7 models we track have a score here.
- 1
Gemini 2.5 Flash Sep 2025 94%
- 2 Claude Sonnet 4.5 91%
- 3
Qwen2.5-72B 90.5%
- 4 GPT-4o mini 86.6%
- 5
InternLM3-8B-Instruct 67.4%
- 6 internlm-chat-20b ↑1 42.8%
- 7 gpt-oss-20b ↓1 33.8%
Unfair Contract Detection (Acc)
7 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 69.7%
- 2
Qwen2.5-72B ↑1 64.5%
- 3
Gemini 2.5 Flash Sep 2025 ↓2 61.5%
- 4 GPT-4o mini 51.3%
- 5
InternLM3-8B-Instruct 42.3%
- 6 gpt-oss-20b 28.2%
- 7 internlm-chat-20b 26.1%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The VLegal-Bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official VLegal-Bench 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
- 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 14 months ago, in September 2025, for Claude Sonnet 4.5. 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 (23 entries).
- Pass Models scored 7 models we track, plus 16 other entries on its leaderboard.
- Warning Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5.
- Note Frontier models None of the top 10 has a score. The strongest tested is Claude Sonnet 4.5 (#81).
- Pass Headroom The best is 72%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 7 rows for models we track and 16 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.