HELM AIR-Bench 2024 - #5.18: Life support: leaderboard
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 1.5 Flash (002) | 93.33 |
| 2 | GPT-OSS-20B | 93.33 |
| 3 | GPT-OSS-120B | 86.67 |
| 4 | O3 (2025-04-16) | 80 |
| 5 | Claude 3 Opus (20240229) | 76.67 |
| 6 | GPT-5 | 73.33 |
| 7 | Claude 3.5 Sonnet (20241022) | 73.33 |
| 8 | Gemini 1.5 Pro (002) | 73.33 |
| 9 | Qwen 3 Next 80B A3B (Thinking) | 73.33 |
| 10 | Claude Haiku 4.5 (20251001) | 73.33 |
| 11 | Claude 3.7 Sonnet (20250219) | 73.33 |
| 12 | Claude Sonnet 4.5 | 66.67 |
| 13 | Claude Sonnet 4 (20250514) | 66.67 |
| 14 | GPT-5 Nano | 63.33 |
| 15 | GPT-3.5 Turbo (0613) | 63.33 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-5-18-life-support · How It Works · Data refreshed daily, snapshot 2026-09-19.