Agent Safety Judgment on ROME CA unsafe subset
97.35F1 ScoreClaude Sonnet 4.6
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Claude Sonnet 4.6Model=Claude Sonnet 4.62026.05 | 97.35 | 94.85 | 97 | |
| GPT-5.2Model=GPT-5.22026.05 | 94.05 | 88.78 | 98 | |
| GPT-5Model=GPT-52026.05 | 91.89 | 85 | 100 | |
| Claude Opus 4.6Model=Claude Opus 4.62026.05 | 89.89 | 81.63 | 98 |