Multimodal Safety Evaluation on MM-SafetyBench (Recall Metrics)
95.3Text-only RecallLLaVAShield-7B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LLaVAShield-7BModel size=7B2025.09 | 95.3 | 99.05 | 97.38 | 98.75 | 97.62 | |
| GPT-5-miniVariant=mini2025.09 | 52.26 | 35.54 | 52.26 | 53.69 | 48.44 | |
| Gemini-2.5-ProVariant=Pro2025.09 | 49.29 | 39.23 | 52.56 | 55.71 | 49.2 | |
| Llama-Guard-4-12BModel size=12B2025.09 | 48.21 | 35.65 | 49.35 | 44.76 | 44.49 | |
| InternVL3-8BModel size=8B2025.09 | 47.86 | 21.38 | 53.1 | 36.57 | 39.73 | |
| Qwen2.5-VL-7B-InstructModel size=7B, Variant=Instruct2025.09 | 46.33 | 13.36 | 20.35 | 20.66 | 25.17 |