HELM AIR-Bench 2024 - #6.22: Legal: leaderboard
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-OSS-120B | 26.67 |
| 2 | Claude Haiku 4.5 (20251001) | 26.67 |
| 3 | GPT-5 Nano | 20 |
| 4 | GPT-OSS-20B | 20 |
| 5 | Claude 3.5 Sonnet (20241022) | 13.33 |
| 6 | Claude Sonnet 4.5 | 10 |
| 7 | Claude 3 Opus (20240229) | 10 |
| 8 | GPT-3.5 Turbo (1106) | 10 |
| 9 | Claude 3.5 Sonnet (20240620) | 6.67 |
| 10 | Claude 3 Haiku (20240307) | 6.67 |
| 11 | Granite 4.0 Micro | 6.67 |
| 12 | Gemini 2.0 Pro (Preview 02-05) | 6.67 |
| 13 | Claude Sonnet 4 (20250514) | 6.67 |
| 14 | Claude Opus 4 (20250514) | 6.67 |
| 15 | Qwen 2 72B Instruct | 3.33 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-6-22-legal · How It Works · Data refreshed daily, snapshot 2026-09-19.