HELM AIR-Bench 2024 - #17: Adult content: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-3 risk category "#17: Adult content", under the level-2 category Sexual Content 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 3.5 Sonnet (20241022) | 96.38 | #302 |
| 2 | Claude 3.5 Sonnet (20240620) | 88.41 | #308 |
| 3 | Claude Haiku 4.5 (20251001) | 87.68 | #264 |
| 4 | Claude Sonnet 4 (20250514) | 86.23 | #223 |
| 5 | Claude 3.7 Sonnet (20250219) | 85.51 | #209 |
| 6 | Claude Opus 4 (20250514) | 84.06 | #164 |
| 7 | Qwen 2.5 72B Instruct | 83.33 | #469 |
| 8 | Granite 3.3 8B Instruct | 83.33 | #815 |
| 9 | GPT-4 Turbo | 82.61 | #417 |
| 10 | Gemini 2.0 Flash | 81.88 | #351 |
| 11 | Granite 4.0 Micro | 81.88 | #866 |
| 12 | GPT-OSS-20B | 81.16 | #523 |
| 13 | GPT-OSS-120B | 80.43 | #346 |
| 14 | Llama 3 8B Instruct | 80.43 | #1116 |
| 15 | Granite 4.0 H Small | 80.43 | #666 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-17-adult-content · How It Works · Data refreshed daily, snapshot 2026-11-05.