Multilingual text query safety on PolyGuardPrompts
88.24F1 ScoreYuFeng-XGuard-Reason-8B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| YuFeng-XGuard-Reason-8BSource=Open-Source Models, Model Scale=8B2026.06 | 88.24 | 90.02 | 89.66 | 86.87 | 86.48 | |
| SingGuard-8BSource=Open-Source Models, Model Scale=8B2026.06 | 88.02 | 89.6 | 87.41 | 88.64 | 88.72 | |
| HaloGuard 1.0Size=4B2026.07 | 88 | — | — | — | — | |
| Qwen3Guard-8B-strictSource=Open-Source Models, Model Scale=8B, Safety Policy=strict2026.06 | 87.83 | 89.59 | 88.58 | 87.1 | 87.12 | |
| SingGuard-4BSource=Open-Source Models, Model Scale=4B2026.06 | 87.59 | 89.48 | 89.13 | 86.1 | 87.3 | |
| GuardReasoner-VL-7BSource=Open-Source Models, Model Scale=7B2026.06 | 87.53 | 89.81 | 92.71 | 82.91 | 80.98 | |
| PolyGuard-QwenSize=7B2026.07 | 87.1 | — | — | — | — | |
| HaloGuard 1.0Size=0.8B2026.07 | 86.1 | — | — | — | — | |
| PolyGuard MinistralSize=8B2026.07 | 86 | — | — | — | — | |
| SingGuard-2BSource=Open-Source Models, Model Scale=2B2026.06 | 85.34 | 87.32 | 85.11 | 85.57 | 85.49 | |
| Qwen3Guard-8B-looseSource=Open-Source Models, Model Scale=8B, Safety Policy=loose2026.06 | 84.07 | 87.72 | 95.4 | 75.14 | 68.03 | |
| PolyGuard SmolSize=0.5B2026.07 | 83.8 | — | — | — | — | |
| GraniteGuardianSource=Open-Source Models2026.06 | 79.91 | 82.79 | 80.46 | 79.38 | 83.71 | |
| GPT-5.1Source=Closed-Source Models2026.06 | 78.77 | 83.69 | 87.65 | 71.53 | 70.07 | |
| Gemini3-ProSource=Closed-Source Models2026.06 | 77.41 | 79.44 | 85.97 | 70.4 | 78.19 | |
| WildGuardSource=Open-Source Models2026.06 | 76.91 | 83.42 | 96.27 | 64.03 | 66.07 | |
| Llama Guard 3Source=Open-Source Models2026.06 | 72.28 | 80.79 | 95.68 | 58.08 | 58.63 | |
| ShieldGemma-9BSource=Open-Source Models, Model Scale=9B2026.06 | 49.82 | 70.43 | 92.9 | 34.04 | 46.07 | |
| Qwen3-VL-235BSource=Open-Source Models, Model Scale=235B2026.06 | 34.02 | 59.55 | 90.22 | 20.96 | 31.38 |