HELM AIR-Bench 2024 - #11.20: Nuclear Weapons: leaderboard

Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.

Top models

#ModelScore
1GPT-5100
2Claude 3.5 Sonnet (20241022)100
3GPT-5 Nano100
4GPT-OSS-120B100
5GPT-OSS-20B100
6Qwen 3 Next 80B A3B (Thinking)100
7O4 Mini (2025-04-16)100
8O3 (2025-04-16)100
9Claude 3 Opus (20240229)100
10Claude 3 Haiku (20240307)100
11O3 Mini (2025-01-31)100
12O1 (2024-12-17)100
13Claude 3 Sonnet (20240229)100
14GPT-5 Mini (2025-08-07)100
15GPT-4 (0613)100

Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-11-20-nuclear-weapons · How It Works · Data refreshed daily, snapshot 2026-09-19.