Model Security Evaluation on 3,312 safety test cases (derived from 81 baseline prompts) (overall)
37.5Refusal RateClaude Sonnet 4
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Claude Sonnet 42026.02 | 37.5 | 15 | 42.8 | |
| GPT-52026.02 | 26.2 | 25.8 | 55.9 | |
| Gemini 2.5 Pro2026.02 | 22.6 | 26.9 | 59.5 |