Agent Safety Judgment on ROME SDM unsafe
100F1 ScoreClaude Sonnet 4.6
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Claude Sonnet 4.6Model=Claude Sonnet 4.62026.05 | 100 | 100 | 97 | |
| GPT-5.2Model=GPT-5.22026.05 | 98.48 | 97 | 100 | |
| Claude Opus 4.6Model=Claude Opus 4.62026.05 | 98.43 | 96.91 | 97 | |
| GPT-5Model=GPT-52026.05 | 93.05 | 87 | 100 |