OOD safety category inference (Stage 2) on OpenAI Moderation
36.45Mean RewardGemini2.5-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini2.5-Flash2025.12 | 36.45 | |
| ProGuard-7B2025.12 | 29.05 | |
| ProGuard-3B2025.12 | 28.31 | |
| GPT4o-mini2025.12 | 8.87 |
Loading the SOTA2 catalog…