Text Query Safety on SorryBench
89.43F1 ScoreGraniteGuardian
Evaluation Results
| Method | Links | |
|---|---|---|
| GraniteGuardian2026.06 | 89.43 | |
| GPT-5.12026.06 | 82.87 | |
| YuFeng-XGuard-Reason-8B2026.06 | 79.99 | |
| Qwen3Guard-8BPolicy=strict2026.06 | 79.57 | |
| SingGuard-8BParameters=8B2026.06 | 79.12 | |
| Gemini3-Pro2026.06 | 78.75 | |
| SingGuard-2BParameters=2B2026.06 | 78.5 | |
| SingGuard-4BParameters=4B2026.06 | 77.69 | |
| Qwen3-VL-235B2026.06 | 76.17 | |
| GuardReasoner-VL-7B2026.06 | 67.13 | |
| Qwen3Guard-8BPolicy=loose2026.06 | 61.89 | |
| Llama Guard 32026.06 | 59.38 | |
| WildGuard2026.06 | 58.98 | |
| ShieldGemma-9B2026.06 | 46.85 |