HELM AIR-Bench 2024 - #14.18: Beliefs: leaderboard
Metric: Refusal Rate (%). Source: crfm.stanford.edu. 87 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.1 8B Instruct | 100 |
| 2 | GPT-5 | 100 |
| 3 | Claude 3.5 Sonnet (20241022) | 100 |
| 4 | Qwen 2.5 72B Instruct | 100 |
| 5 | Claude Sonnet 4.5 | 100 |
| 6 | Claude 3.5 Sonnet (20240620) | 100 |
| 7 | GPT-5 Nano | 100 |
| 8 | GPT-OSS-120B | 100 |
| 9 | Llama 4 Scout Instruct | 100 |
| 10 | GPT-4o Mini (2024-07-18) | 100 |
| 11 | GPT-4.1 (2025-04-14) | 100 |
| 12 | Gemini 2.5 Pro (Preview 03-25) | 100 |
| 13 | GPT-4o (2024-08-06) | 100 |
| 14 | GPT-OSS-20B | 100 |
| 15 | Qwen 3 Next 80B A3B (Thinking) | 100 |
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-14-18-beliefs · How It Works · Data refreshed daily, snapshot 2026-09-19.