HELM AIR-Bench 2024 - #16.35: Vulgarity: leaderboard
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 100 |
| 2 | Qwen 3 Next 80B A3B (Thinking) | 100 |
| 3 | Claude 3 Haiku (20240307) | 100 |
| 4 | GPT-OSS-20B | 94.44 |
| 5 | Claude Haiku 4.5 (20251001) | 94.44 |
| 6 | Claude 3 Sonnet (20240229) | 94.44 |
| 7 | GPT-5 Nano | 88.89 |
| 8 | GPT-4 Turbo | 88.89 |
| 9 | Claude 3 Opus (20240229) | 88.89 |
| 10 | Granite 3.3 8B Instruct | 88.89 |
| 11 | Claude Sonnet 4.5 | 83.33 |
| 12 | Claude 3.5 Sonnet (20240620) | 83.33 |
| 13 | Gemini 2.0 Flash | 83.33 |
| 14 | Llama 4 Maverick Instruct FP8 | 83.33 |
| 15 | Claude 3.7 Sonnet (20250219) | 83.33 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-16-35-vulgarity · How It Works · Data refreshed daily, snapshot 2026-09-19.