BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

LegalBench

Legal reasoning tasks written by lawyers. Run by Vals AI.

As of 25 Oct 2026, Claude Fable 5 leads LegalBench on BenchLeader with 88.6%, ahead of Claude Fable 5.1 at 88.5%, across 143 model configurations with a published result.

Published by
Vals AI
Category
Knowledge
Index weight
0.5
Models
143
Data as of
25 Oct 2026
Newest result
13 Oct 2026

Vals AI (vals.ai).

What the test looks like

LegalBench is a collection of legal reasoning tasks written by lawyers and legal scholars, from identifying contract clauses to applying rules to facts.

How it is scored

Accuracy across tasks, run by Vals AI.

What to keep in mind

Tasks are US-centric and mostly classification, so it measures legal reading more than drafting or advice.

Top 15
  1. 1Claude Fable 588.6%
  2. 2Claude Fable 5.188.5%
  3. 3Gemini 4 Argon (high)88.3%
  4. 4Muse Wonder Full (max)87.4%
  5. 5Gemini 3.1 Pro (high)87.4%
  6. 6Gemini 3.7 Flash (high)87.3%
  7. 7Gemini 3 Pro (high)87.0%
  8. 8Gemini 3.8 Flash (high)87.0%
  9. 9GPT-5.6 Sol (max)87.0%
  10. 10Claude Opus 587.0%
  11. 11Gemini 3 Flash (high)86.9%
  12. 12Gemini 3.6 Flash (high)86.7%
  13. 13GPT-5.5 (xhigh)86.5%
  14. 14Grok 4.6 (high)86.3%
  15. 15GLM-Legal (max)86.2%
143 of 143
#
1Claude Fable 588.6%66.52026-10-13
2Claude Fable 5.188.5%66.42026-10-13
3Gemini 4 Argonhigh88.3%70.72026-10-13
4Muse Wonder Fullmax87.4%70.62026-10-13
5Gemini 3.1 Prohigh87.4%60.32026-10-13
6Gemini 3.7 Flashhigh87.3%63.52026-10-13
7Gemini 3 Prohigh87.0%60.02026-10-13
8Gemini 3.8 Flashhigh87.0%64.42026-10-13
9GPT-5.6 Solmax87.0%68.02026-10-13
10Claude Opus 587.0%65.12026-10-13
11Gemini 3 Flashhigh86.9%58.32026-10-13
12Gemini 3.6 Flashhigh86.7%62.02026-10-13
13GPT-5.5xhigh86.5%66.12026-10-13
14Grok 4.6high86.3%62.92026-10-13
15GLM-Legalmax86.2%–2026-10-13
16Kimi K3max86.2%64.82026-10-13
17GPT-5.4xhigh86.0%64.72026-10-13
18GPT-5high86.0%57.92026-10-13
19Grok 4.5high86.0%59.32026-10-13
20GPT-5.1high85.7%57.72026-10-13
21MiniMax M385.4%56.42026-10-13
22Claude Opus 4.6thinking85.3%62.52026-10-13
23Muse Spark 1.2xhigh85.3%63.22026-10-13
24Claude Opus 4.785.3%63.22026-10-13
25GPT-5.6 Terramax85.1%64.42026-10-13
26Qwen3.5 Plusthinking85.1%57.32026-10-13
27Muse Spark 1.1xhigh85.0%61.62026-10-13
28Qwen3.7max84.9%60.62026-10-13
29GLM 5.3max84.8%63.52026-10-13
30Kimi K2.684.7%55.32026-10-13
31Claude Opus 4.5thinking84.6%59.92026-10-13
32Grok 4.384.5%54.62026-10-13
33Grok 4.7xhigh84.4%61.62026-10-13
34GLM 5.184.4%55.92026-10-13
35Gemini 2.5 Pro84.3%53.92026-10-13
36Qwen3.5 Flash84.3%52.02026-10-13
37Qwen3.6 Plus84.2%58.82026-10-13
38Muse Spark84.2%64.32026-10-13
39Claude Sonnet 4.5thinking84.1%52.62026-10-13
40GLM 5.284.1%51.82026-10-13
41Gemini 3.5 Flash Litehigh84.1%47.92026-10-13
42GLM-5thinking84.1%58.92026-10-13
43GPT-5.6 Lunamax84.0%59.92026-10-13
44MiniMax M2.784.0%55.62026-10-13
45GLM 5.3 Flashmax83.9%54.22026-10-13
46Claude Sonnet 583.9%55.92026-10-13
47Gemini 2.5 Flash83.8%52.72026-10-13
48o3high83.8%53.92026-10-13
49Gemini 3.1 Flash Litehigh83.8%48.82026-10-13
50Qwen3.8 Maxmax83.6%63.32026-10-13
51Gemini 3.5 Flashhigh83.6%63.12026-10-13
52Claude Opus 4.883.6%60.12026-10-13
53Claude Opus 4.183.5%51.52026-10-13
54GLM 4.783.4%51.92026-10-13
55DeepSeek V4.1 Flashhigh83.3%55.62026-10-13
56Grok 483.2%58.02026-10-13
57Grok 3 Mini Fasthigh83.1%53.72026-10-13
58GPT-4.1high83.1%46.72026-10-13
59Claude Opus 483.1%52.62026-10-13
60Mercury 2.5high83.1%–2026-10-13

More knowledge benchmarks · All LLM benchmarks

Cite as: BenchLeader, “LegalBench leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/vals_legalbench.

LegalBench: questions

What does LegalBench measure?
LegalBench is a collection of legal reasoning tasks written by lawyers and legal scholars, from identifying contract clauses to applying rules to facts. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads LegalBench?
Claude Fable 5 leads LegalBench with 88.6% as of 25 Oct 2026, ahead of Claude Fable 5.1 at 88.5%.
How many models have LegalBench results?
143 model configurations have a LegalBench result on BenchLeader, all taken from Vals AI.
Who runs LegalBench and how often is it updated?
LegalBench is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
Does LegalBench count toward the BenchLeader Index?
Yes. LegalBench contributes to the knowledge category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.