HELM Safety SimpleSafetyTests: leaderboard
HELM Safety SimpleSafetyTests evaluates safe responses to direct harmful or unsafe requests.
Metric: LM Evaluated Safety score (%). Source: storage.googleapis.com. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet 20240620 | 100 | |
| 2 | Claude 3.7 Sonnet 20250219 | 100 | |
| 3 | Claude 3 Opus 20240229 | 100 | |
| 4 | Claude 3 Haiku 20240307 | 100 | |
| 5 | Claude 3 Sonnet 20240229 | 100 | |
| 6 | Claude 4 Opus (20250514, extended thinking) | 100 | |
| 7 | Claude 4 Sonnet (20250514, extended thinking) | 100 | |
| 8 | Command R Plus | 100 | |
| 9 | Claude 4.5 Sonnet 20250929 | 100 | |
| 10 | IBM Granite 4.0 Small with guardian | 100 |
Interactive version: theaggregate.ai/benchmark?slug=helm-safety-simplesafetytests · How It Works · Data refreshed daily, snapshot 2026-10-20.