HELM AIR-Bench 2024 - #13: Harassment: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-3 risk category "#13: Harassment", under the level-2 category Hate/Toxicity 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. Saturation forecast: Not forecast. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 98.66 | #302 |
| 2 | Claude Haiku 4.5 (20251001) | 97.85 | #264 |
| 3 | GPT-OSS-20B | 96.24 | #523 |
| 4 | GPT-OSS-120B | 95.43 | #346 |
| 5 | Claude 3 Haiku (20240307) | 94.09 | #780 |
| 6 | Claude Sonnet 4.5 | 93.82 | #148 |
| 7 | Claude Sonnet 4 (20250514) | 91.94 | #223 |
| 8 | Claude 3 Sonnet (20240229) | 91.4 | #694 |
| 9 | Granite 3.3 8B Instruct | 90.32 | #815 |
| 10 | GPT-5.1 (2025-11-13) | 89.78 | #107 |
| 11 | Claude Opus 4 (20250514) | 89.25 | #164 |
| 12 | Claude 3.5 Sonnet (20240620) | 88.71 | #308 |
| 13 | GPT-5 Nano | 87.37 | #439 |
| 14 | Claude 3 Opus (20240229) | 84.68 | #428 |
| 15 | Granite 4.0 H Small | 83.87 | #666 |
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-2024-13-harassment · How It Works · Data refreshed daily, snapshot 2026-11-05.