HELM AIR-Bench 2024 - #11: Weapon Usage & Development: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-3 risk category "#11: Weapon Usage & Development", under the level-2 category Violence & Extremism of the AIR 2024 regulation-derived taxonomy. 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 | O3 (2025-04-16) | 100 | #125 |
| 2 | GPT-OSS-120B | 98.89 | #346 |
| 3 | Claude 3 Haiku (20240307) | 98.89 | #780 |
| 4 | Claude 3.5 Sonnet (20241022) | 97.22 | #302 |
| 5 | O1 (2024-12-17) | 96.67 | #157 |
| 6 | Claude 3 Sonnet (20240229) | 96.67 | #694 |
| 7 | Claude 3 Opus (20240229) | 96.67 | #428 |
| 8 | O4 Mini (2025-04-16) | 95.56 | #183 |
| 9 | GPT-OSS-20B | 94.44 | #523 |
| 10 | GPT-5 | 93.33 | #105 |
| 11 | GPT-5 Nano | 93.33 | #439 |
| 12 | Qwen 3 Next 80B A3B (Thinking) | 93.33 | #312 |
| 13 | GPT-5 Mini (2025-08-07) | 93.33 | #175 |
| 14 | Claude 3.5 Sonnet (20240620) | 91.11 | #308 |
| 15 | GPT-5.1 (2025-11-13) | 91.11 | #107 |
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-2024-11-weapon-usage-development · How It Works · Data refreshed daily, snapshot 2026-11-05.