HELM AIR-Bench 2024 - #20.8: Pornography: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#20.8: Pornography", under level-3 "#20: Monetized" 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 Haiku (20240307) | 100 | #780 |
| 2 | Claude 3 Sonnet (20240229) | 86.67 | #694 |
| 3 | Claude 3.5 Sonnet (20241022) | 80 | #302 |
| 4 | Qwen 3 Next 80B A3B (Thinking) | 80 | #312 |
| 5 | Claude Haiku 4.5 (20251001) | 73.33 | #264 |
| 6 | Llama 3 8B Instruct | 66.67 | #1116 |
| 7 | Claude 3 Opus (20240229) | 66.67 | #428 |
| 8 | GPT-OSS-20B | 60 | #523 |
| 9 | Llama 4 Maverick Instruct FP8 | 60 | #424 |
| 10 | Granite 4.0 Micro | 60 | #866 |
| 11 | Gemini 2.0 Pro (Preview 02-05) | 60 | #318 |
| 12 | Granite 3.3 8B Instruct | 60 | #815 |
| 13 | Granite 4.0 H Small | 56.67 | #666 |
| 14 | GPT-OSS-120B | 46.67 | #346 |
| 15 | Gemini 2.0 Flash | 46.67 | #351 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-20-8-pornography · How It Works · Data refreshed daily, snapshot 2026-11-05.