HELM AIR-Bench 2024 - #6: Advice in Heavily Regulated Industries: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-3 risk category "#6: Advice in Heavily Regulated Industries", under the level-2 category Operational Misuses of the AIR 2024 regulation-derived taxonomy. Every prompt is a harmful request; a GPT-4o judge grades each response from full compliance (0) to refusal (1), and the mean is shown as 0-100, where higher means more refusals (safer).
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 (20251001) | 40.67 | #264 |
| 2 | Claude 3.5 Sonnet (20241022) | 29.33 | #302 |
| 3 | GPT-3.5 Turbo (1106) | 26 | #952 |
| 4 | Claude Sonnet 4.5 | 24.67 | #148 |
| 5 | GPT-OSS-20B | 24.67 | #523 |
| 6 | GPT-3.5 Turbo (0613) | 24.67 | #1106 |
| 7 | GPT-OSS-120B | 22.67 | #346 |
| 8 | GPT-5 Nano | 22 | #439 |
| 9 | O1 (2024-12-17) | 22 | #157 |
| 10 | O3 Mini (2025-01-31) | 20.67 | #249 |
| 11 | Granite 4.0 Micro | 20.67 | #866 |
| 12 | GPT-3.5 Turbo (0125) | 18 | #975 |
| 13 | Gemini 1.5 Flash (002) | 17.33 | #587 |
| 14 | Claude 3 Opus (20240229) | 17.33 | #428 |
| 15 | Claude 3 Haiku (20240307) | 16.67 | #780 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-6-advice-in-heavily-regulated-industries · How It Works · Data refreshed daily, snapshot 2026-11-05.