Benchmark Almanac

VLegal-Bench

Gemini 2.5 Flash Sep 2025 72% leads since December 2025
Not in model cards

Models answer questions and solve legal scenarios under Vietnamese law. Legal experts check that reference answers follow authoritative documents.

Released
December 2025
Kind
Law
Level
University
Scored
Checked automatically
Answers
Open-ended
Tasks
All public
Measures
Capability
Subject
Law
Language
Vietnamese
In and out
Text → Text
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
23
Entries in its paper, 7 models we track

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:

  1. 1 Gemini 2.5 Flash Sep 2025, Zero-shot without reasoning 72%
  2. 2 Qwen2.5-32B-Instruct Zero-shot without reasoning 71.3%
  3. 3 gpt-4o (snapshot unspecified) Zero-shot without reasoning 70%
  4. 4 Claude Sonnet 4.5 Zero-shot without reasoning 69.8%
  5. 5 Qwen2.5-14B-Instruct Zero-shot without reasoning 68.6%
  6. 6 Qwen2.5-72B Zero-shot without reasoning 68%
  7. 7 CMC-Legal-32B Zero-shot without reasoning 66.2%
  8. 8 GPT-4o mini Zero-shot without reasoning 65.2%
  9. 9 Qwen3-4b-legal-pretrain Zero-shot without reasoning 62.8%
  10. 10 SeaLLMs-v3-7B-Chat Zero-shot without reasoning 62.2%
  11. 11 gemma-2-27b-it Zero-shot without reasoning 57.9%
  12. 12 InternLM3-8B-Instruct Zero-shot without reasoning 57.2%
  13. 13 Llama-3.1-8B-Instruct Zero-shot without reasoning 55.9%
  14. 14 Llama-3.1-70B-Instruct Zero-shot without reasoning 55.7%
  15. 15 gemma-2-9b-it Zero-shot without reasoning 52.9%
  16. 16 Qwen2.5-3B-Instruct Zero-shot without reasoning 52.8%
  17. 17 Qwen2.5-7B-Instruct Zero-shot without reasoning 50.4%
  18. 18 SeaLLMs-v3-1.5B-Chat Zero-shot without reasoning 47.7%
  19. 19 BloomVN-8B-chat Zero-shot without reasoning 46.7%
  20. 20 Llama-2-7b-chat-hf Zero-shot without reasoning 25.5%
  21. 21 Llama-2-13b-chat-hf Zero-shot without reasoning 22.7%
  22. 22 gpt-oss-20b Zero-shot without reasoning 22.7%
  23. 23 internlm-chat-20b Zero-shot without reasoning 16.4%

Source: The VLegal-Bench paper

#EntryScoreSettingDateSource
1 Gemini 2.5 Flash Sep 2025 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 Qwen2.5-72B 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 InternLM3-8B-Instruct 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
#ModelScoreSettingDateSource
1 Gemini 2.5 Flash Sep 2025 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 Qwen2.5-72B 68% Zero-shot without reasoning Sep 2024 Official leaderboard
4 GPT-4o mini 65.2% Zero-shot without reasoning Jul 2024 Official leaderboard
5 InternLM3-8B-Instruct 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. 1 Gemini 2.5 Flash Sep 2025 72%
  2. 2 Claude Sonnet 4.5 69.8%
  3. 3 Qwen2.5-72B 68%
  4. 4 GPT-4o mini 65.2%
  5. 5 InternLM3-8B-Instruct 57.2%
  6. 6 gpt-oss-20b 22.7%
  7. 7 internlm-chat-20b 16.4%

Number one over time

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

Google DeepMind
Benchmark paper Not tested since Dec 2025

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.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Gemini 2.5 Flash Qwen2.5-72B internlm-chat-20b 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Gemini 2.5 Flash GPT-4o mini

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

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 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.

How we check