Multimodal Safety on VLSBench
99.39F1 ScoreLLaVAShield
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVAShieldCategory=Open-Source Models2026.06 | 99.39 | |
| SingGuard-8BCategory=Our Models2026.06 | 96.68 | |
| SingGuard-2BCategory=Our Models2026.06 | 96.03 | |
| SingGuard-4BCategory=Our Models2026.06 | 95.33 | |
| GPT-5.1Category=Closed-Source Models2026.06 | 88.26 | |
| Qwen3-VL-235BCategory=Open-Source Models2026.06 | 87.51 | |
| Gemini3-ProCategory=Closed-Source Models2026.06 | 86.82 | |
| SafeGuard-VLCategory=Open-Source Models2026.06 | 80.8 | |
| Qwen3-VL-8BCategory=Open-Source Models2026.06 | 75.84 | |
| LlavaGuardCategory=Open-Source Models2026.06 | 68.72 | |
| Qwen3-VL-4BCategory=Open-Source Models2026.06 | 68.68 | |
| GuardReasoner-VL-7BCategory=Open-Source Models2026.06 | 63.06 | |
| ShieldGemma-2Category=Open-Source Models2026.06 | 62.85 | |
| LlamaGuard4Category=Open-Source Models2026.06 | 11.6 | |
| LlamaGuard3-Vision-11BCategory=Open-Source Models2026.06 | 7.31 |