Deceptive Behavior Evaluation on Apollo Research Deception Evaluations
28.36Covert Action Rategpt-5-thinking
Evaluation results
| Method | Links | |
|---|---|---|
| gpt-5-thinkingVariant=helpful-only2025.12 | 28.36 | |
| OpenAI o3Variant=Default2025.12 | 8.24 | |
| gpt-5-thinkingVariant=Default2025.12 | 3.97 |