87.4%
#1
LegalBench accuracy of Gemini 3.1 Pro Preview, the highest of the top 15; GLM 4.7 was 15th at 83.4%
What the chart shows
LegalBench is a crowd-sourced benchmark of tasks that mirror legal work: spotting issues, recalling and applying rules, predicting outcomes and interpreting legal text. Gemini 3.1 Pro Preview led at 87.4%, followed by Gemini 3 Pro (87%) and Gemini 3 Flash (86.9%).
All of the top 15 score above 83%. The spread from first to 15th (GLM 4.7, 83.4%) is 4.0 percentage points, a range the AI Index says makes the models hard to tell apart.
Key numbers
| Measure | Value |
|---|---|
| Gemini 3.1 Pro Preview (02/26) | 87.40% |
| Gemini 3 Pro (11/25) | 87.04% |
| Gemini 3 Flash (12/25) | 86.86% |
| GPT 5 | 86.02% |
| GLM 4.7, 15th | 83.36% |
| Gap, 1st to 15th | 4.04 percentage points |
How to read it
- The benchmark covers six types of legal reasoning: issue spotting, rule recall, outcome prediction, rule application, interpretation and rhetorical understanding.
- Scores are from Vals.ai’s LegalBench leaderboard; the AI Index shows the top 15 models.
Source: Vals.ai, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.5.13: LegalBench: accuracy. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion