HELM AIR-Bench 2024 - #7: Supporting Malicious Organized Groups: leaderboard
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 98.89 |
| 2 | Claude 3.5 Sonnet (20240620) | 98.89 |
| 3 | GPT-OSS-20B | 98.89 |
| 4 | Claude Haiku 4.5 (20251001) | 98.89 |
| 5 | Claude Sonnet 4.5 | 97.78 |
| 6 | GPT-OSS-120B | 97.78 |
| 7 | Llama 3 8B Instruct | 97.78 |
| 8 | O3 (2025-04-16) | 97.78 |
| 9 | Claude 3 Sonnet (20240229) | 97.78 |
| 10 | Granite 3.3 8B Instruct | 97.78 |
| 11 | GPT-5 | 96.67 |
| 12 | Qwen 3 Next 80B A3B (Thinking) | 96.67 |
| 13 | O4 Mini (2025-04-16) | 96.67 |
| 14 | GPT-5 Nano | 95.56 |
| 15 | Claude 3 Opus (20240229) | 95.56 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-7-supporting-malicious-organized-groups · How It Works · Data refreshed daily, snapshot 2026-09-19.