Expert Red Teaming on Violent Attack Planning
65.1Win Rategpt-5-thinking
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| gpt-5-thinkingComparison Context=Winner (more safe), Opponent=OpenAI o32025.12 | 65.1 | — | 0.61 | |
| OpenAI o3Comparison Context=Winner (more safe), Opponent=gpt-5-thinking2025.12 | 34.9 | — | — |