HELM AIR-Bench 2024 - #16.36: Derogatory comments: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#16.36: Derogatory comments", under level-3 "#16: Offensive Language" and level-2 "Hate/Toxicity" of the regulation-derived AIR 2024 taxonomy. The category holds only 24 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. Saturation forecast: Not forecast. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Granite 4.0 H Small | 100 | #666 |
| 2 | Claude 3.5 Sonnet (20241022) | 95.83 | #302 |
| 3 | Claude Haiku 4.5 (20251001) | 95.83 | #264 |
| 4 | Claude 3.7 Sonnet (20250219) | 95.83 | #209 |
| 5 | Claude Sonnet 4.5 | 93.75 | #148 |
| 6 | GPT-OSS-20B | 93.75 | #523 |
| 7 | GPT-4 Turbo | 91.67 | #417 |
| 8 | Claude Sonnet 4 (20250514) | 91.67 | #223 |
| 9 | Claude Opus 4 (20250514) | 91.67 | #164 |
| 10 | Granite 3.3 8B Instruct | 91.67 | #815 |
| 11 | GPT-OSS-120B | 89.58 | #346 |
| 12 | Claude 3.5 Sonnet (20240620) | 87.5 | #308 |
| 13 | Claude 3 Haiku (20240307) | 87.5 | #780 |
| 14 | Granite 4.0 Micro | 85.42 | #866 |
| 15 | GPT-4.1 Mini | 83.33 | #363 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-16-36-derogatory-comments · How It Works · Data refreshed daily, snapshot 2026-11-05.