Skip to main content
benchlm.ai

Vals LegalBench (LegalBench)

Vals AI legal benchmark with issue, rule, conclusion, interpretation, and rhetoric task views.

How BenchLM shows LegalBench

BenchLM mirrors the public Vals AI LegalBench leaderboard captured from https://www.vals.ai/benchmarks/legal_bench and updated by Vals on May 16, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

LegalBench is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

108 Vals rows6 task viewspublic datasetTasks: Overall, Issue Tasks, Rule Tasks, Conclusion Tasks, Interpretation TasksDisplay only
LegalBench on Vals AIVals methodologyVals home

LegalBench score on LegalBench — May 16, 2026

BenchLM mirrors the published legalbench score view for LegalBench. Gemini 3.1 Pro Preview leads the public snapshot at 87.40% , followed by Gemini 3 Pro Preview (87.03%) and Gemini 3 Flash Preview (86.86%). BenchLM does not use these results to rank models overall.

1

Gemini 3.1 Pro Preview

Google

google/gemini-3.1-pro-preview

87.40%
Overall —
2

Gemini 3 Pro Preview

Google

google/gemini-3-pro-preview

87.03%
Overall —
3

Gemini 3 Flash Preview

Google

google/gemini-3-flash-preview

86.86%
Overall —
108 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated May 16, 2026

The published LegalBench snapshot is tightly clustered at the top: Gemini 3.1 Pro Preview sits at 87.40%, while the third row is only 0.54 points behind. The broader top-10 spread is 2.29 points, so many of the published scores sit in a relatively narrow band.

108 models have been evaluated on LegalBench. The benchmark falls in the External benchmark mirrors category. BenchLM tracks this category separately from its weighted global scoring system, so these results are best compared on the dedicated Korean benchmark views. LegalBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About LegalBench

Year

2026

Tasks

Legal reasoning task views

Format

Accuracy score

Difficulty

Professional legal reasoning

BenchLM mirrors Vals LegalBench as display-only legal-domain evidence and does not use it in weighted rankings.

LegalBenchPublic benchmark source

BenchLM freshness & provenance

Version

LegalBench 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

LegalBench score table (108 models)

1
Gemini 3.1 Pro Previewgoogle/gemini-3.1-pro-preview
87.40%
2
Gemini 3 Pro Previewgoogle/gemini-3-pro-preview
87.03%
3
Gemini 3 Flash Previewgoogle/gemini-3-flash-preview
86.86%
4
GPT-5.5openai/gpt-5.5
86.52%
5
GPT-5.4openai/gpt-5.4-2026-03-05
86.04%
6
GPT-5openai/gpt-5-2025-08-07
86.02%
7
GPT-5.1openai/gpt-5.1-2025-11-13
85.68%
8
Claude Opus 4.6 Thinkinganthropic/claude-opus-4-6-thinking
85.30%
9
Claude Opus 4.7anthropic/claude-opus-4-7
85.25%
10
Qwen3.5 Plus Thinkingalibaba/qwen3.5-plus-thinking
85.10%
11
Kimi K2.6 Thinkingkimi/kimi-k2.6-thinking
84.74%
12
Claude Opus 4.5 20251101 Thinkinganthropic/claude-opus-4-5-20251101-thinking
84.60%
13
Grok 4.3grok/grok-4.3
84.46%
14
GLM 5.1 Thinkingzai/glm-5.1-thinking
84.39%
15
Gemini 2.5 Pro Exp 03 25google/gemini-2.5-pro-exp-03-25
84.32%
16
Qwen3.5 Flashalibaba/qwen3.5-flash
84.28%
17
Qwen3.6 Plusalibaba/qwen3.6-plus
84.23%
18
Muse Sparkmeta/muse_spark
84.22%
19
Claude Sonnet 4.5 20250929 Thinkinganthropic/claude-sonnet-4-5-20250929-thinking
84.08%
20
GLM 5 Thinkingzai/glm-5-thinking
84.06%
21
MiniMax M2.7minimax/MiniMax-M2.7
83.98%
22
Gemini 2.5 Flash Preview 04 17google/gemini-2.5-flash-preview-04-17
83.80%
23
Gemini 3.1 Flash Lite Previewgoogle/gemini-3.1-flash-lite-preview
83.76%
24
O3openai/o3-2025-04-16
83.76%
25
Gemini 3.5 Flashgoogle/gemini-3.5-flash
83.60%
26
Claude Opus 4.1anthropic/claude-opus-4-1-20250805
83.46%
27
GLM 4.7zai/glm-4.7
83.36%
28
Grok 4 0709grok/grok-4-0709
83.19%
29
Grok 3 Mini Fast High Reasoninggrok/grok-3-mini-fast-high-reasoning
83.14%
30
GPT-4.1openai/gpt-4.1-2025-04-14
83.10%
31
Claude Opus 4anthropic/claude-opus-4-20250514
83.07%
32
Claude Sonnet 4anthropic/claude-sonnet-4-20250514
82.95%
33
Claude Opus 4.5anthropic/claude-opus-4-5-20251101
82.84%
34
GPT-5.2openai/gpt-5.2-2025-12-11
82.76%
35
Gemini 2.5 Flash Preview 09 2025 Thinkinggoogle/gemini-2.5-flash-preview-09-2025-thinking
82.63%
36
Grok 3grok/grok-3
82.59%
37
Claude 3.7 Sonnet 20250219 Thinkinganthropic/claude-3-7-sonnet-20250219-thinking
82.52%
38
Gemini 2.5 Flash Preview 09 2025google/gemini-2.5-flash-preview-09-2025
82.49%
39
Grok 4.1 Fast Reasoninggrok/grok-4-1-fast-reasoning
82.45%
40
GPT-4oopenai/gpt-4o-2024-11-20
82.21%
41
Claude Sonnet 4.6anthropic/claude-sonnet-4-6
82.12%
42
Claude Sonnet 4 20250514 Thinkinganthropic/claude-sonnet-4-20250514-thinking
82.06%
43
Gemini 2.5 Flash Lite Preview 09 2025 Thinkinggoogle/gemini-2.5-flash-lite-preview-09-2025-thinking
82.04%
44
Grok 3 Mini Fast Low Reasoninggrok/grok-3-mini-fast-low-reasoning
81.92%
45
Qwen3 Maxalibaba/qwen3-max
81.86%
46
GPT-5 Miniopenai/gpt-5-mini-2025-08-07
81.77%
47
Moonshotai Kimi K2 Instructtogether/moonshotai/Kimi-K2-Instruct
81.45%
48
Claude Haiku 4.5 20251001 Thinkinganthropic/claude-haiku-4-5-20251001-thinking
81.24%
49
DeepSeek V3fireworks/deepseek-v3
80.76%
50
Grok 4 Fast Reasoninggrok/grok-4-fast-reasoning
80.60%
51
GPT-4 Turboopenai/gpt-4-turbo
80.46%
52
O1openai/o1-2024-12-17
80.39%
53
Qwen3 Max Previewalibaba/qwen3-max-preview
80.33%
54
DeepSeek V4 Prodeepseek/deepseek-v4-pro
80.32%
55
Kimi K2 Thinkingkimi/kimi-k2-thinking
80.20%
56
Qwen3 235b A22bfireworks/qwen3-235b-a22b
80.18%
57
GPT-4oopenai/gpt-4o-2024-08-06
80.12%
58
Claude 3.7 Sonnetanthropic/claude-3-7-sonnet-20250219
80.00%
59
MiniMax M2.5 Lightningminimax/MiniMax-M2.5-Lightning
79.96%
60
Command A 03 2025cohere/command-a-03-2025
79.70%
61
GLM 4.6zai/glm-4.6
79.61%
62
Qwen Qwen2.5 72B Instruct Turbotogether/Qwen/Qwen2.5-72B-Instruct-Turbo
79.40%
63
O4 Miniopenai/o4-mini-2025-04-16
79.19%
64
Mistral Large 2512mistralai/mistral-large-2512
79.14%
65
Grok 4.1 Fast Non Reasoninggrok/grok-4-1-fast-non-reasoning
79.11%
66
Gemini 2.5 Flash Lite Preview 09 2025google/gemini-2.5-flash-lite-preview-09-2025
79.01%
67
Grok 4 Fast Non Reasoninggrok/grok-4-fast-non-reasoning
78.40%
68
Gemini 2.0 Flash 001google/gemini-2.0-flash-001
78.36%
69
GPT-4.1 Miniopenai/gpt-4.1-mini-2025-04-14
78.04%
70
GPT-5.4 Nanoopenai/gpt-5.4-nano-2026-03-17
77.92%
71
Llama4 Maverick Instruct Basicfireworks/llama4-maverick-instruct-basic
77.81%
72
Grok 4.20 0309 Reasoninggrok/grok-4.20-0309-reasoning
77.74%
73
DeepSeek V3 0324fireworks/deepseek-v3-0324
77.73%
74
Meta Llama Llama 3.3 70B Instruct Turbotogether/meta-llama/Llama-3.3-70B-Instruct-Turbo
77.18%
75
DeepSeek V3p2 Thinkingfireworks/deepseek-v3p2-thinking
76.08%
76
GPT Oss 120bfireworks/gpt-oss-120b
75.94%
77
GLM 4.5zai/glm-4.5
75.63%
78
MiniMax M2.1minimax/MiniMax-M2.1
75.45%
79
Jamba 1.5 Largeai21labs/jamba-1.5-large
74.16%
80
Meta Llama Llama 4 Scout 17B 16E Instructtogether/meta-llama/Llama-4-Scout-17B-16E-Instruct
72.04%
81
Jamba Large 1.6ai21labs/jamba-large-1.6
71.96%
82
O3 Miniopenai/o3-mini-2025-01-31
71.54%
83
GPT Oss 20bfireworks/gpt-oss-20b
70.85%
84
Claude 3.5 Haikuanthropic/claude-3-5-haiku-20241022
70.33%
85
Gemini 1.0 Pro 002google/gemini-1.0-pro-002
70.25%
86
Jamba Mini 1.6ai21labs/jamba-mini-1.6
69.73%
87
Qwen Qwen2.5 7B Instruct Turbotogether/Qwen/Qwen2.5-7B-Instruct-Turbo
69.56%
88
Mistral Small 2503mistralai/mistral-small-2503
69.16%
89
Gemini 1.5 Pro 002google/gemini-1.5-pro-002
69.08%
90
Command R Pluscohere/command-r-plus
68.29%
91
Google Gemma 2 9b Ittogether/google/gemma-2-9b-it
68.14%
92
Google Gemma 2 27b Ittogether/google/gemma-2-27b-it
67.98%
93
DeepSeek R1fireworks/deepseek-r1
67.32%
94
Jamba 1.5 Miniai21labs/jamba-1.5-mini
66.62%
95
DeepSeek V3p2fireworks/deepseek-v3p2
65.96%
96
GPT-3.5 Turboopenai/gpt-3.5-turbo
64.37%
97
Gemini 1.5 Flash 002google/gemini-1.5-flash-002
63.24%
98
Meta Llama Llama 2 70b Hftogether/meta-llama/Llama-2-70b-hf
62.43%
99
Mistral Medium 2505mistralai/mistral-medium-2505
61.98%
100
GPT-4.1 Nanoopenai/gpt-4.1-nano-2025-04-14
61.06%
101
Mistralai Mixtral 8x7B V0.1together/mistralai/Mixtral-8x7B-v0.1
55.84%
102
Magistral Medium 2509mistralai/magistral-medium-2509
54.77%
103
Mistralai Mistral 7B V0.1together/mistralai/Mistral-7B-v0.1
53.77%
104
Togethercomputer Llama 2 13btogether/togethercomputer/llama-2-13b
53.70%
105
Togethercomputer Llama 2 7btogether/togethercomputer/llama-2-7b
51.87%
106
GPT-5 Nanoopenai/gpt-5-nano-2025-08-07
50.13%
107
Magistral Small 2509mistralai/magistral-small-2509
40.02%
108
Command Rcohere/command-r
32.97%

FAQ

What does LegalBench measure?

Vals AI legal benchmark with issue, rule, conclusion, interpretation, and rhetoric task views.

Which model leads the published LegalBench snapshot?

Gemini 3.1 Pro Preview currently leads the published LegalBench snapshot with 87.40% legalbench score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on LegalBench?

108 AI models are included in BenchLM's mirrored LegalBench snapshot, based on the public leaderboard captured on May 16, 2026.

Last updated: May 16, 2026 · mirrored from the public benchmark leaderboard

The AI models change fast. We track them for you.

For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.