HELM AIR-Bench 2024 - #6.23: Medical/Pharmaceutical: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#6.23: Medical/Pharmaceutical", under level-3 "#6: Advice in Heavily Regulated Industries" and level-2 "Operational Misuses". The category holds only 15 prompts, so scores move in coarse steps. A GPT-4o judge grades each response from compliance (0) to refusal (1); the mean is shown as 0-100, higher meaning more refusals (safer).
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 60 | #302 |
| 2 | GPT-OSS-20B | 60 | #523 |
| 3 | Gemini 1.5 Flash (002) | 36.67 | #587 |
| 4 | Claude Sonnet 4.5 | 33.33 | #148 |
| 5 | GPT-OSS-120B | 33.33 | #346 |
| 6 | Claude Haiku 4.5 (20251001) | 33.33 | #264 |
| 7 | GPT-3.5 Turbo (0613) | 30 | #1106 |
| 8 | Gemini 2.0 Pro (Preview 02-05) | 26.67 | #318 |
| 9 | GPT-3.5 Turbo (1106) | 26.67 | #952 |
| 10 | Gemini 1.5 Pro (002) | 20 | #368 |
| 11 | Claude Sonnet 4 (20250514) | 20 | #223 |
| 12 | Granite 4.0 Micro | 16.67 | #866 |
| 13 | GPT-4 (0613) | 16.67 | #493 |
| 14 | GPT-4o (2024-08-06) | 13.33 | #343 |
| 15 | Qwen 3 Next 80B A3B (Thinking) | 13.33 | #312 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-6-23-medical-pharmaceutical · How It Works · Data refreshed daily, snapshot 2026-11-05.