Text-based safety moderation on Aegis
84F1 ScoreOMNIGUARD-7B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OMNIGUARD-7BSize=7B2025.12 | 84 | 84.1 | |
| Qwen3-235BSize=235B2025.12 | 83.3 | 83.6 | |
| OMNIGUARD-3BSize=3B2025.12 | 82.2 | 82.6 | |
| Qwen2.5-Omni-7BSize=7B2025.12 | 77.3 | 73.9 | |
| Qwen2.5-7BSize=7B2025.12 | 75.4 | 77.5 | |
| ThinkGuardSize=8B2025.12 | 69.9 | 74.6 | |
| Qwen2.5-72BSize=72B2025.12 | 68.9 | 73.2 | |
| LLaMA Guard 3Size=8B2025.12 | 65.8 | 73.5 | |
| LLaMA-3.3-70BSize=70B2025.12 | 65.7 | 69.4 | |
| LLaMA Guard 2Size=8B2025.12 | 59 | 69.2 | |
| GPT-4oSize=-2025.12 | 54.3 | 65.6 | |
| LLaMA Guard 1Size=7B2025.12 | 53 | 66.9 |