HELM Lite - LegalBench - PROA: leaderboard
HELM Lite v1.13.0 LegalBench PROA task: decide whether a statutory provision creates a private right of action (yes/no); 95 items per model, exact match. 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 | Llama 3.1 70B Instruct | 100 | #567 |
| 2 | Mistral Large 2 (Jul) | 100 | #490 |
| 3 | Gemini 1.5 Pro (001) | 100 | #465 |
| 4 | Claude 3.5 Sonnet (20241022) | 98.95 | #302 |
| 5 | GPT-4o (2024-05-13) | 98.95 | #337 |
| 6 | Gemini 2.0 Flash (Preview) | 98.95 | #376 |
| 7 | Claude 3 Opus (20240229) | 98.95 | #428 |
| 8 | Llama 3.2 90B Vision Instruct | 98.95 | #621 |
| 9 | GPT-4 Turbo (Preview) | 98.95 | #416 |
| 10 | Jamba 1.5 Large | 98.95 | #677 |
| 11 | Llama 3.3 70B Instruct | 97.89 | #541 |
| 12 | Qwen 2.5 72B Instruct | 97.89 | #469 |
| 13 | Gemini 1.5 Flash (002) | 97.89 | #587 |
| 14 | GPT-4o (2024-08-06) | 97.89 | #343 |
| 15 | Gemma 2 27B (IT) | 97.89 | #652 |
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-proa · How It Works · Data refreshed daily, snapshot 2026-11-05.