Skip to content

Benchmarks · Vals AI

LegalBench (Vals leaderboard)

Note: Vals has no benchmark aimed at Dutch or EU law. Measures legal reasoning on short, well-delimited tasks (open dataset by Guha et al.; Vals runs every model with the same settings). Split into issue, rule, conclusion, interpretation and rhetoric tasks. The top ten sit between 87.0 and 88.6 percent (as at 13 October 2026), so the ranking separates little. The original dataset (https://github.com/HazyResearch/legalbench) carries a licence per task; its maintainers ask users to follow the dataset creator's licence.

Models measured
113
Top
88.56 %Claude Fable 5within the standard error of ±0.33
Version
1
Measured
13 October 2026
Go to the sourceThis benchmark does not count towards the PrudentScore.

01

Ranking

At the top: Top 15

Vals AI Legal Research Bench, all-pass [1]. Hatched bars are rejected by the compliance gate.

  1. 1Claude Fable 588.56%
  2. 2Claude Fable 5.188.51%
  3. 3Gemini 4 Argon88.30%
  4. 4Wonder Full87.42%
  5. 5Gemini 3.1 Pro Preview (02/26)87.40%
  6. 6Gemini 3.7 Flash87.26%
  7. 7Gemini 3 Pro (11/25)87.03%
  8. 8Gemini 3.8 Flash86.99%
  9. 9Claude Opus 586.97%
  10. 10GPT-5.6 Sol86.97%
  11. 11Gemini 3 Flash (12/25)86.86%
  12. 12Gemini 3.6 Flash86.70%
  13. 13GPT 5.586.52%
  14. 14Grok 4.686.31%
  15. 15GLM Legal86.23%

Source: Vals AI (LegalBench (Vals leaderboard)) · Licence: CC BY 4.0 · Adaptation: Prudai selected the models, translated the explanations, added the gate verdict and the PrudentScore and rounded figures to two decimals.

Passes the compliance gateRejected by the compliance gate
LegalBench (Vals leaderboard) 1
# ModelScoreCost/taskGatePrudentScore
1Claude Fable 5Anthropic88.56%± 0.33-Rejectedno score
2Claude Fable 5.1Anthropic88.51%± 0.42-Rejectedno score
3Gemini 4 ArgonGoogle88.30%± 0.37-Passes79
4Wonder FullMeta87.42%± 0.41-Rejectedno score
5Gemini 3.1 Pro Preview (02/26)Google87.40%± 0.33-Rejectedno score
6Gemini 3.7 FlashGoogle87.26%± 0.42-Passes58
7Gemini 3 Pro (11/25)Google87.03%± 0.37-Rejectedno score
8Gemini 3.8 FlashGoogle86.99%± 0.43-Passes62
9Claude Opus 5Anthropic86.97%± 0.42-Passes72
10GPT-5.6 SolOpenAI86.97%± 0.41-Passes57
11Gemini 3 Flash (12/25)Google86.86%± 0.36-Rejectedno score
12Gemini 3.6 FlashGoogle86.70%± 0.41-Passes46
13GPT 5.5OpenAI86.52%± 0.41-Passes51
14Grok 4.6xAI (SpaceXAI)86.31%± 0.42-Rejectedno score
15GLM LegalZ.ai86.23%± 0.41-Rejectedno score
16Kimi K3Moonshot AI86.21%± 0.41-Rejectedno score
17GPT 5.4 (xhigh)OpenAI86.04%± 0.41-Passes0
18GPT 5OpenAI86.02%± 0.38-Passesno score
19Grok 4.5xAI (SpaceXAI)85.97%± 0.41-Rejectedno score
20GPT 5.1OpenAI85.68%± 0.40-Passesno score
21MiniMax-M3MiniMax85.42%± 0.44-Rejectedno score
22Claude Opus 4.6 (Thinking)Anthropic85.30%± 0.37-Passesno score
23Muse Spark 1.2Meta85.26%± 0.45-Rejectedno score
24Claude Opus 4.7Anthropic85.25%± 0.45-Passes54
25GPT-5.6 TerraOpenAI85.11%± 0.45-Passes53
26Qwen 3.5 PlusAlibaba85.10%± 0.41-Not yet assessedno score
27Muse Spark 1.1Meta84.98%± 0.42-Rejectedno score
28Qwen 3.7 MaxAlibaba84.91%± 0.41-Not yet assessedno score
29GLM 5.3Z.ai84.84%± 0.40-Rejectedno score
30Kimi K2.6Moonshot AI84.74%± 0.45-Rejectedno score
31Claude Opus 4.5 (Thinking)Anthropic84.60%± 0.39-Passesno score
32Grok 4.3xAI (SpaceXAI)84.46%± 0.42-Rejectedno score
33GLM 5.1Z.ai84.39%± 0.43-Rejectedno score
34Grok 4.7xAI (SpaceXAI)84.39%± 0.46-Rejectedno score
35Qwen 3.5 FlashAlibaba84.28%± 0.41-Not yet assessedno score
36Qwen 3.6 PlusAlibaba84.23%± 0.42-Not yet assessedno score
37Muse SparkMeta84.22%± 0.42-Rejectedno score
38Claude Sonnet 4.5 (Thinking)Anthropic84.08%± 0.45-Passesno score
39=Gemini 3.5 Flash LiteGoogle84.07%± 0.44-Passes38
39=GLM 5.2Z.ai84.07%± 0.45-Rejectedno score
41GLM 5Z.ai84.06%± 0.41-Passesno score
42GPT-5.6 LunaOpenAI84.03%± 0.42-Passes56
43MiniMax-M2.7MiniMax83.98%± 0.41-Rejectedno score
44GLM 5.3 FlashZ.ai83.93%± 0.43-Rejectedno score
45Claude Sonnet 5Anthropic83.92%± 0.46-Passes59
46Gemini 3.1 Flash Lite PreviewGoogle83.76%± 0.41-Not yet assessedno score
47o3OpenAI83.76%± 0.42-Passesno score
48Hy4 PreviewTencent83.76%± 0.41-Rejectedno score
49Qwen 3.8 MaxAlibaba83.61%± 0.42-Not yet assessedno score
50Gemini 3.5 FlashGoogle83.60%± 0.85-Passes47
51Claude Opus 4.8Anthropic83.57%± 0.47-Passes62
52Claude Opus 4.1 (Nonthinking)Anthropic83.46%± 0.45-Rejectedno score
53GLM 4.7Z.ai83.36%± 0.47-Passesno score
54DeepSeek V4.1 FlashDeepSeek83.28%± 0.46-Rejectedno score
55Grok 4xAI (SpaceXAI)83.19%± 0.47-Rejectedno score
56GPT 4.1OpenAI83.10%± 0.43-Passesno score
57Claude Opus 4 (Nonthinking)Anthropic83.07%± 0.44-Rejectedno score
58Mercury 2.5Inception Labs83.06%± 0.48-Not yet assessedno score
59Inkling SmallThinking Machines82.99%± 0.46-Rejectedno score
60Claude Sonnet 4 (Nonthinking)Anthropic82.95%± 0.41-Rejectedno score
61InklingThinking Machines82.95%± 0.50-Rejectedno score
62Claude Opus 4.5 (Nonthinking)Anthropic82.84%± 0.43-Passesno score
63GPT 5.2OpenAI82.76%± 0.40-Passesno score
64Gemini 2.5 Flash Preview (9/25) (Thinking)Google82.63%± 0.41-Rejectedno score
65Gemini 2.5 Flash Preview (9/25) (Nonthinking)Google82.49%± 0.43-Rejectedno score
66Grok 4.1 Fast (Reasoning)xAI (SpaceXAI)82.45%± 0.46-Rejectedno score
67Qwen 3.8 27BAlibaba82.43%± 0.47-Rejectedno score
68DeepSeek V4 Pro (0813)DeepSeek82.36%± 0.44-Rejectedno score
69GPT 4o (2024-11-20)OpenAI82.21%± 0.46-Passesno score
70Claude Sonnet 4.6Anthropic82.12%± 0.48-Passes55
71Nemotron 3 UltraNVIDIA82.07%± 0.46-Rejectedno score
72Claude Sonnet 4 (Thinking)Anthropic82.06%± 0.46-Rejectedno score
73Gemini 2.5 Flash Lite (9/25) (Thinking)Google82.04%± 0.44-Rejectedno score
74Qwen 3 MaxAlibaba81.86%± 0.40-Not yet assessedno score
75GPT 5 MiniOpenAI81.77%± 0.45-Passesno score
76Kimi K2 InstructMoonshot AI81.45%± 0.44-Rejectedno score
77Claude Haiku 4.5 (Thinking)Anthropic81.24%± 0.49-Passes32
78Grok 4 Fast (Reasoning)xAI (SpaceXAI)80.60%± 0.45-Rejectedno score
79Qwen 3 Max PreviewAlibaba80.33%± 0.40-Not yet assessedno score
80DeepSeek V4 Pro (2026-04-23)DeepSeek80.32%± 0.47-Rejectedno score
81Kimi K2 ThinkingMoonshot AI80.20%± 0.46-Rejectedno score
82MiniMax-M2.5MiniMax79.96%± 0.47-Passesno score
83Command ACohere79.70%± 0.42-Rejectedno score
84Ling 3.0 FlashAnt Group79.69%± 0.52-Rejectedno score
85GLM 4.6Z.ai79.61%± 0.52-Rejectedno score
86o4 MiniOpenAI79.19%± 0.50-Passesno score
87Mistral Large 3Mistral AI79.14%± 0.46-Passesno score
88Grok 4.1 Fast Non-ReasoningxAI (SpaceXAI)79.11%± 0.46-Rejectedno score
89Gemini 2.5 Flash Lite (9/25) (Nonthinking)Google79.01%± 0.44-Rejectedno score
90MiMo V2.5Xiaomi78.89%± 0.51-Rejectedno score
91Grok 4 Fast (Non-Reasoning)xAI (SpaceXAI)78.40%± 0.47-Rejectedno score
92GPT 4.1 MiniOpenAI78.04%± 0.42-Passesno score
93GPT 5.4 NanoOpenAI77.92%± 0.43-Passes29
94Llama 4 MaverickMeta77.81%± 0.42-Rejectedno score
95Grok 4.20 (Reasoning)xAI (SpaceXAI)77.74%± 0.48-Rejectedno score
96DeepSeek V4 Flash 0731DeepSeek77.71%± 0.50-Rejectedno score
97MiMo V2.5 ProXiaomi77.13%± 0.55-Rejectedno score
98DeepSeek V3.2 (Thinking)DeepSeek76.08%± 0.49-Passesno score
99GPT OSS 120BOpenAI75.94%± 0.52-Passesno score
100GLM 4.5Z.ai75.63%± 0.51-Rejectedno score
101MiniMax-M2.1MiniMax75.45%± 0.54-Passesno score
102Laguna M.1Poolside75.14%± 0.59-Rejectedno score
103Llama 4 ScoutMeta72.04%± 0.48-Rejectedno score
104Laguna XS.2Poolside71.03%± 0.56-Rejectedno score
105GPT OSS 20BOpenAI70.85%± 0.61-Passesno score
106Mistral Small 3.1 (03/2025)Mistral AI69.16%± 0.51-Passesno score
107DeepSeek V3.2 (Nonthinking)DeepSeek65.96%± 0.42-Passesno score
108Mistral Medium 3.1 (05/2025)Mistral AI61.98%± 0.47-Passesno score
109Command A+Cohere61.42%± 0.63-Rejectedno score
110GPT 4.1 NanoOpenAI61.06%± 0.59-Passesno score
111Magistral Medium 1.2 (09/2025)Mistral AI54.77%± 0.56-Not yet assessedno score
112GPT 5 NanoOpenAI50.13%± 0.47-Passesno score
113Magistral Small 1.2 (09/2025)Mistral AI40.02%± 0.47-Passesno score

Source: Vals AI (LegalBench (Vals leaderboard)) · Licence: CC BY 4.0 · Adaptation: Prudai selected the models, translated the explanations, added the gate verdict and the PrudentScore and rounded figures to two decimals.

Vals AI · Version 1 · Measured 13 October 2026 · Go to the source. The PrudentScore is a composition by Prudai B.V. based on publicly published figures and does not represent the view or endorsement of Vals AI, Harvey or Artificial Analysis. Figures from Artificial Analysis appear in running text only, with attribution and a link, as that party's terms require [9].

Other benchmarks

Claude and Anthropic, GPT and OpenAI, Gemini and Google, Grok and xAI, Kimi and Moonshot AI, Muse and Meta, Mistral, DeepSeek, GLM and Z.ai, Harvey, Vals AI and Artificial Analysis are trademarks of their respective owners. They are named for identification only and this implies no partnership, sponsorship or endorsement.