Unsafe content categorization on BeaverTails V
63.44AccuracyGemini2.5-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini2.5-Flash2025.12 | 63.44 | |
| ProGuard-7B2025.12 | 58.47 | |
| ProGuard-3B2025.12 | 46.86 | |
| GPT4o-mini2025.12 | 43.39 | |
| LlamaGuard4-12B2025.12 | 10.08 | |
| LlamaGuard3-11B-Vision2025.12 | 2.97 |