HELM AIR-Bench 2024 - #17.2: Suggestive: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#17.2: Suggestive", under level-3 "#17: Adult content" 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) | 83.33 | #302 |
| 2 | Llama 3.1 8B Instruct | 76.67 | #1017 |
| 3 | Granite 3.3 8B Instruct | 76.67 | #815 |
| 4 | Claude Haiku 4.5 (20251001) | 73.33 | #264 |
| 5 | Granite 4.0 H Small | 73.33 | #666 |
| 6 | Granite 4.0 Micro | 73.33 | #866 |
| 7 | Claude Sonnet 4 (20250514) | 73.33 | #223 |
| 8 | GPT-5 Nano | 70 | #439 |
| 9 | Mistral 7B Instruct (v0.3) | 70 | #1255 |
| 10 | Qwen 3 Next 80B A3B (Thinking) | 70 | #312 |
| 11 | Llama 4 Maverick Instruct FP8 | 70 | #424 |
| 12 | Claude Sonnet 4.5 | 66.67 | #148 |
| 13 | Claude 3.5 Sonnet (20240620) | 66.67 | #308 |
| 14 | Llama 4 Scout Instruct | 66.67 | #547 |
| 15 | Llama 3 70B Instruct | 66.67 | #640 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-17-2-suggestive · How It Works · Data refreshed daily, snapshot 2026-11-05.