HELM Safety: leaderboard
HELM Safety: Measures model robustness, truthfulness, calibration, bias, harmfulness, jailbreak resistance, or alignment-relevant behavior.
Metric: Mean score (imported). Source: crfm.stanford.edu. Source: benchmarklist.com. 59 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | o3 (2025-04-16) | 98.16 | |
| 2 | Claude 4 Sonnet (20250514, extended thinking) | 98.07 | |
| 3 | Claude 3.5 Sonnet (20240620) | 97.67 | #299 |
| 4 | o1 (2024-12-17) | 97.58 | |
| 5 | Claude 4 Sonnet (20250514) | 97.44 | |
| 6 | o4-mini (2025-04-16) | 97.32 | |
| 7 | Claude 4 Opus (20250514, extended thinking) | 96.94 | |
| 8 | Claude 4 Opus (20250514) | 96.77 | |
| 9 | Claude 3 Opus (20240229) | 96.75 | #412 |
| 10 | GPT-4.5 (2025-02-27 preview) | 96.47 |
Interactive version: theaggregate.ai/benchmark?slug=helm-safety · How It Works · Data refreshed daily, snapshot 2026-10-20.