HELM Lite - LegalBench: leaderboard
HELM Lite v1.13.0 LegalBench score: the mean exact match over the five LegalBench legal-reasoning tasks HELM runs (Abercrombie, corporate lobbying, function of decision section, international citizenship questions and PROA). Percent correct, 0-100, higher is better.
Metric: Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 91 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 1.5 Pro (001) | 75.72 | #465 |
| 2 | Gemini 1.5 Pro (002) | 74.7 | #368 |
| 3 | Qwen 2.5 72B Instruct | 74.01 | #469 |
| 4 | Nova Pro | 73.62 | #594 |
| 5 | GPT-4o (2024-05-13) | 73.32 | #337 |
| 6 | Llama 3 70B | 73.29 | #632 |
| 7 | GPT-4 Turbo | 72.73 | #417 |
| 8 | Llama 3.3 70B Instruct | 72.54 | #541 |
| 9 | GPT-4o (2024-08-06) | 72.09 | #343 |
| 10 | DeepSeek V3 | 71.78 | #329 |
| 11 | GPT-4 (0613) | 71.28 | #493 |
| 12 | Qwen 2 72B Instruct | 71.17 | #579 |
| 13 | Mixtral 8x22B | 70.79 | #725 |
| 14 | Llama 3.1 405B Instruct | 70.71 | #470 |
| 15 | Claude 3.5 Sonnet (20240620) | 70.66 | #308 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-lite-legalbench · How It Works · Data refreshed daily, snapshot 2026-11-05.