HELM AIR-Bench 2024 - #11: Weapon Usage & Development: leaderboard
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 (2025-04-16) | 100 |
| 2 | GPT-OSS-120B | 98.89 |
| 3 | Claude 3 Haiku (20240307) | 98.89 |
| 4 | Claude 3.5 Sonnet (20241022) | 97.22 |
| 5 | Claude 3 Opus (20240229) | 96.67 |
| 6 | O1 (2024-12-17) | 96.67 |
| 7 | Claude 3 Sonnet (20240229) | 96.67 |
| 8 | O4 Mini (2025-04-16) | 95.56 |
| 9 | GPT-OSS-20B | 94.44 |
| 10 | GPT-5 | 93.33 |
| 11 | GPT-5 Nano | 93.33 |
| 12 | Qwen 3 Next 80B A3B (Thinking) | 93.33 |
| 13 | GPT-5 Mini (2025-08-07) | 93.33 |
| 14 | Claude 3.5 Sonnet (20240620) | 91.11 |
| 15 | Claude Haiku 4.5 (20251001) | 90 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-11-weapon-usage-development · How It Works · Data refreshed daily, snapshot 2026-09-19.