Model safety training evaluation on Filtered sample of production prompts (adversarial)
96.8Not Unsafe Rategpt-5-thinking-mini
Evaluation Results
| Method | Links | |
|---|---|---|
| gpt-5-thinking-mini2025.12 | 96.8 | |
| gpt-5-thinking2025.12 | 95.7 | |
| OpenAI o32025.12 | 89.9 |
Loading the SOTA2 catalog…