HELM AIR-Bench: leaderboard
HELM AIR-Bench: Measures model robustness, truthfulness, calibration, bias, harmfulness, jailbreak resistance, or alignment-relevant behavior.
Metric: Refusal Rate (%). Source: crfm.stanford.edu. Source: storage.googleapis.com. Saturation forecast: Estimated already saturated. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 (20251001) | 93.2 | #261 |
| 2 | Claude 3.5 Sonnet (20241022) | 90.8 | #291 |
| 3 | Claude Sonnet 4.5 | 89.8 | #142 |
| 4 | Claude Sonnet 4 (20250514) | 88.3 | #213 |
| 5 | GPT-OSS-120B | 88 | #341 |
| 6 | GPT-5 Nano | 87.8 | #425 |
| 7 | GPT-5 | 87.7 | #98 |
| 8 | Qwen 3 Next 80B A3B (Thinking) | 86.7 | #320 (Qwen 3 Next 80B A3B) |
| 9 | GPT-5.1 (2025-11-13) | 86.2 | #101 |
| 10 | GPT-OSS-20B | 86 | #518 |
| 11 | Claude 3.5 Sonnet (20240620) | 85.9 | #299 |
| 12 | GPT-5 Mini (2025-08-07) | 85.7 | #172 |
| 13 | Claude Opus 4 (20250514) | 85.7 | #154 |
| 14 | Claude 3 Sonnet (20240229) | 84.7 | #677 |
| 15 | O3 (2025-04-16) | 84.5 | #121 |
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 · How It Works · Data refreshed daily, snapshot 2026-10-20.