HELM AIR-Bench 2024 - #18.5: Erotic chats: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#18.5: Erotic chats", under level-3 "#18: Erotic" and level-2 "Sexual Content" of the regulation-derived AIR 2024 taxonomy. 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) | 100 | #302 |
| 2 | Claude Haiku 4.5 (20251001) | 100 | #264 |
| 3 | Claude Sonnet 4 (20250514) | 100 | #223 |
| 4 | Claude Opus 4 (20250514) | 100 | #164 |
| 5 | Llama 3.1 8B Instruct | 86.67 | #1017 |
| 6 | Claude 3 Sonnet (20240229) | 86.67 | #694 |
| 7 | Claude 3.7 Sonnet (20250219) | 83.33 | #209 |
| 8 | Granite 4.0 H Small | 83.33 | #666 |
| 9 | Claude Sonnet 4.5 | 80 | #148 |
| 10 | Claude 3.5 Sonnet (20240620) | 80 | #308 |
| 11 | Claude 3 Haiku (20240307) | 80 | #780 |
| 12 | Llama 3 8B Instruct | 76.67 | #1116 |
| 13 | Granite 4.0 Micro | 76.67 | #866 |
| 14 | GPT-4 Turbo | 70 | #417 |
| 15 | Llama 4 Maverick Instruct FP8 | 70 | #424 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-18-5-erotic-chats · How It Works · Data refreshed daily, snapshot 2026-11-05.