Multimodal Safety on JailBreakV
99.61F1 ScoreLLaVAShield
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVAShieldCategory=Open-Source Models2026.06 | 99.61 | |
| GPT-5.1Category=Closed-Source Models2026.06 | 98.24 | |
| SingGuard-8BCategory=Our Models2026.06 | 97.28 | |
| Gemini3-ProCategory=Closed-Source Models2026.06 | 97.27 | |
| SingGuard-4BCategory=Our Models2026.06 | 96.94 | |
| SingGuard-2BCategory=Our Models2026.06 | 95.05 | |
| GuardReasoner-VL-7BCategory=Open-Source Models2026.06 | 94.52 | |
| Qwen3-VL-235BCategory=Open-Source Models2026.06 | 93.75 | |
| SafeGuard-VLCategory=Open-Source Models2026.06 | 93.19 | |
| Qwen3-VL-8BCategory=Open-Source Models2026.06 | 92.84 | |
| Qwen3-VL-4BCategory=Open-Source Models2026.06 | 92.06 | |
| ShieldGemma-2Category=Open-Source Models2026.06 | 90.17 | |
| LlamaGuard4Category=Open-Source Models2026.06 | 85.88 | |
| LlamaGuard3-Vision-11BCategory=Open-Source Models2026.06 | 79.82 | |
| LlavaGuardCategory=Open-Source Models2026.06 | 31.9 |