HELM AIR-Bench 2024 - Political Usage: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-2 risk category "Political Usage" of the AIR 2024 taxonomy, which is derived from government regulations and company AI policies. 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. Saturation forecast: Estimated already saturated. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 (20251001) | 97.33 | #264 |
| 2 | Claude 3 Sonnet (20240229) | 97.33 | #694 |
| 3 | Claude 3.5 Sonnet (20241022) | 97.2 | #302 |
| 4 | GPT-OSS-20B | 96.8 | #523 |
| 5 | Claude 3 Haiku (20240307) | 95.87 | #780 |
| 6 | GPT-OSS-120B | 95.73 | #346 |
| 7 | Claude 3 Opus (20240229) | 94.27 | #428 |
| 8 | O3 (2025-04-16) | 94 | #125 |
| 9 | Claude Sonnet 4.5 | 92.8 | #148 |
| 10 | GPT-5 Mini (2025-08-07) | 92.4 | #175 |
| 11 | Claude 3.5 Sonnet (20240620) | 91.33 | #308 |
| 12 | Qwen 3 Next 80B A3B (Thinking) | 90.67 | #312 |
| 13 | O4 Mini (2025-04-16) | 90.13 | #183 |
| 14 | GPT-5 Nano | 89.07 | #439 |
| 15 | O1 (2024-12-17) | 89.07 | #157 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-political-usage · How It Works · Data refreshed daily, snapshot 2026-11-05.