Multimodal Safety Evaluation on VLGuard (test)
86.78AccuracyLLaVAShield-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLaVAShield-7BModel size=7B2025.09 | 86.78 | 98.7 | 83.64 | 90.55 | |
| GPT-5-miniVariant=mini2025.09 | 84.47 | 76.8 | 98.71 | 86.39 | |
| Gemini-2.5-ProVariant=Pro2025.09 | 78.69 | 71.2 | 94.18 | 81.09 | |
| Llama-Guard-4-12BModel size=12B2025.09 | 66.56 | 48.1 | 99.59 | 64.87 | |
| Qwen2.5-VL-7B-InstructModel size=7B, Variant=Instruct2025.09 | 50.77 | 23.3 | 100 | 37.79 | |
| InternVL3-8BModel size=8B2025.09 | 43.26 | 11.6 | 100 | 20.79 |