Safety Evaluation on PKU-SafeRLHF
93.08F1 ScoreQwen3Guard-8B-loose
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3Guard-8B-looseAccess=Open-Source, Configuration=loose2026.06 | 93.08 | — | |
| SingGuard-4BAccess=Open-Source2026.06 | 92.69 | — | |
| SingGuard-8BAccess=Open-Source2026.06 | 92.67 | — | |
| GuardReasoner-VL-7BAccess=Open-Source2026.06 | 92.54 | — | |
| Qwen3-VL-235BAccess=Open-Source2026.06 | 92.28 | — | |
| YuFeng-XGuard-Reason-8BAccess=Open-Source2026.06 | 91.44 | — | |
| SingGuard-2BAccess=Open-Source2026.06 | 91.39 | — | |
| Qwen3Guard-8B-strictAccess=Open-Source, Configuration=strict2026.06 | 90.43 | — | |
| Llama Guard 3Access=Open-Source2026.06 | 89.08 | — | |
| GPT-5.1Access=Closed-Source2026.06 | 88.8 | — | |
| Gemini3-ProAccess=Closed-Source2026.06 | 86.05 | — | |
| ShieldGemma-9BAccess=Open-Source2026.06 | 82.14 | — | |
| WildGuardAccess=Open-Source2026.06 | 81.89 | — | |
| GraniteGuardianAccess=Open-Source2026.06 | 68.69 | — | |
| ARES (Ministral)Training Configuration=Ours (Ministral)2026.04 | — | 94 | |
| ARES (Qwen)Training Configuration=Ours (Qwen)2026.04 | — | 96 | |
| General Safe-RLHFTraining Configuration=General Safe-RLHF2026.04 | — | 82 | |
| Initial RLHFTraining Configuration=Initial RLHF2026.04 | — | 74 | |
| Original ModelTraining Configuration=Original Model2026.04 | — | 69 |