Multimodal Safety on BeaverTails V
91.54F1 ScoreShieldGemma-2
Evaluation Results
| Method | Links | |
|---|---|---|
| ShieldGemma-2Category=Open-Source Models2026.06 | 91.54 | |
| SingGuard-4BCategory=Our Models2026.06 | 83.37 | |
| SingGuard-8BCategory=Our Models2026.06 | 83.15 | |
| SingGuard-2BCategory=Our Models2026.06 | 80.47 | |
| Qwen3-VL-235BCategory=Open-Source Models2026.06 | 79.49 | |
| GPT-5.1Category=Closed-Source Models2026.06 | 77.25 | |
| Qwen3-VL-4BCategory=Open-Source Models2026.06 | 72.28 | |
| Gemini3-ProCategory=Closed-Source Models2026.06 | 67.21 | |
| Qwen3-VL-8BCategory=Open-Source Models2026.06 | 65.07 | |
| LLaVAShieldCategory=Open-Source Models2026.06 | 64.11 | |
| SafeGuard-VLCategory=Open-Source Models2026.06 | 57.16 | |
| LlavaGuardCategory=Open-Source Models2026.06 | 55.04 | |
| GuardReasoner-VL-7BCategory=Open-Source Models2026.06 | 54.2 | |
| LlamaGuard4Category=Open-Source Models2026.06 | 51.79 | |
| LlamaGuard3-Vision-11BCategory=Open-Source Models2026.06 | 34.38 |