Multimodal Understanding on MMBench OOD
99.84AccuracyVLMShield
Evaluation Results
| Method | Links | |
|---|---|---|
| VLMShieldBase VLM=LLaVA-1.5-13B2026.04 | 99.84 | |
| VLMShieldBase VLM=Qwen2.5-VL-7B-Instruct2026.04 | 99.84 | |
| VLMGuardBase VLM=Qwen2.5-VL-7B-Instruct2026.04 | 98 | |
| ASTRABase VLM=LLaVA-1.5-13B2026.04 | 97.66 | |
| CIDERBase VLM=LLaVA-1.5-13B2026.04 | 97.46 | |
| CIDERBase VLM=Qwen2.5-VL-7B-Instruct2026.04 | 97.46 | |
| VLMGuardBase VLM=LLaVA-1.5-13B2026.04 | 96.92 | |
| ECSOBase VLM=Qwen2.5-VL-7B-Instruct2026.04 | 95.07 | |
| JailGuardBase VLM=Qwen2.5-VL-7B-Instruct2026.04 | 95 | |
| ASTRABase VLM=Qwen2.5-VL-7B-Instruct2026.04 | 94.64 | |
| ECSOBase VLM=LLaVA-1.5-13B2026.04 | 92.8 | |
| JailGuardBase VLM=LLaVA-1.5-13B2026.04 | 91.25 | |
| MirrorCheckBase VLM=LLaVA-1.5-13B2026.04 | 90.17 | |
| MirrorCheckBase VLM=Qwen2.5-VL-7B-Instruct2026.04 | 90.17 |