Safety Moderation on Aegis, StrongReject, PolyGuard Aggregate 2.0
86.4Average F1 ScoreYuFeng-XGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| YuFeng-XGuardSize=8B, Task=s-cls2026.05 | 86.4 | 2.63 | |
| GPT-OSS-SafeGuardSize=20B, Task=s-cls2026.05 | 83.3 | 2.43 | |
| WildGuardSize=7B, Task=s-cls2026.05 | 80.9 | 2.47 | |
| NemotronGuardV2Size=8B, Task=s-cls2026.05 | 78.6 | 2.39 | |
| GLiGuard OmniSize=209M, Task=s-cls+IE2026.05 | 76.9 | 2.78 | |
| Llama3GuardSize=8B, Task=s-cls2026.05 | 75.3 | 2.29 | |
| GLiGuard bi-encSize=145M, Task=s-cls+IE2026.05 | 74.3 | 2.74 | |
| GLiGuard uni-encSize=147M, Task=s-cls+IE2026.05 | 73.8 | 2.72 | |
| Llama4GuardSize=12B, Task=s-cls2026.05 | 69.7 | 2.08 | |
| GLiNER2 Multi v1Size=209M, Task=cls+IE2026.05 | 66.6 | 2.41 | |
| ShieldGemmaSize=9B, Task=s-cls2026.05 | 66.5 | 2.01 | |
| GLiClass Base v1.0Size=150M, Task=cls2026.05 | 64.9 | 2.39 | |
| Longformer-harmful-roSize=149M, Task=s-cls2026.05 | 47.9 | 1.76 | |
| PromptGuard 2*Size=86M, Task=s-cls2026.05 | 35.7 | 1.35 | |
| DeBERTa-v3-base-PISize=184M, Task=p-inj2026.05 | 32.5 | 1.18 |