Safety Evaluation on MM-SafetyBench
0Average ASRRAI
Evaluation Results
| Method | Links | |||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RAIModel=Qwen3-VL-8B2026.02 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | — | — | — | — | — | — | — | — | |
| CBModel=LLaVA-v1.6-Mistral-7B2026.04 | 0 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DefenSeeTarget Model=Llava 22025.12 | 0.03 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdaShield-ATarget Model=Llava 22025.12 | 0.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdaShield-STarget Model=Llava 22025.12 | 0.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CrossGuardCategory=MLLM Guardrails2025.10 | 0.38 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SafeSteerModels=Qwen3-VL2026.05 | 0.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MLLM-ProtectorModels=Qwen3-VL2026.05 | 1.32 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DefenSeeTarget Model=MiniGPT42025.12 | 1.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SafeSteerModels=Qwen2.5-VL2026.05 | 1.75 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BeamModel=LLaVA-v1.6-Mistral-7B2026.04 | 1.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| shiftDCModel=Qwen3-VL-8B2026.02 | 2.24 | 0.31 | 0.62 | 0.93 | 0 | 4.77 | 2.73 | 2.75 | 4.13 | 6.61 | 0.73 | 1.2 | 1.47 | 3.33 | 4.58 | 3.96 | 0.39 | 0.48 | 1.36 | — | — | — | — | — | — | — | — | |
| MRDModels=Qwen3-VL2026.05 | 2.39 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SafetyModel=LLaVA-v1.6-Mistral-7B2026.04 | 2.49 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VanillaModels=Qwen3-VL2026.05 | 2.73 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ECSOModels=Qwen3-VL2026.05 | 2.85 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdashieldModel=Qwen3-VL-8B2026.02 | 3.36 | 0.46 | 0.93 | 1.39 | 0 | 7.16 | 4.09 | 4.13 | 6.19 | 9.91 | 1.1 | 1.8 | 2.21 | 5 | 6.88 | 5.94 | 0.59 | 0.72 | 2.05 | — | — | — | — | — | — | — | — | |
| RAIModel=LLaVA-1.6-7B2026.02 | 3.62 | 0 | 0 | 10.36 | 0 | 12.8 | 3 | 2.45 | 7.25 | 2 | 0 | 11.2 | 0.33 | 4.15 | 0 | 6.83 | 0.65 | 0.21 | 3.97 | — | — | — | — | — | — | — | — | |
| RAIModel=DeepSeek-VL-7B2026.02 | 4.32 | 2.05 | 6.28 | 7.1 | 0.35 | 7.65 | 4.52 | 0.56 | 3.65 | 2.81 | 1.22 | 5.73 | 3.94 | 2.76 | 7.01 | 6.69 | 1.59 | 7.65 | 6.26 | — | — | — | — | — | — | — | — | |
| CoCAModel=Qwen3-VL-8B2026.02 | 6.73 | 0.93 | 1.85 | 2.78 | 0 | 14.32 | 8.18 | 8.25 | 12.38 | 19.82 | 2.21 | 3.6 | 4.42 | 10 | 13.75 | 11.87 | 1.17 | 1.43 | 4.09 | — | — | — | — | — | — | — | — | |
| ECSOModel=Qwen3-VL-8B2026.02 | 6.88 | 0.95 | 1.9 | 2.84 | 0 | 14.64 | 8.36 | 8.44 | 12.66 | 20.26 | 2.25 | 3.68 | 4.52 | 10.22 | 14.06 | 12.13 | 1.2 | 1.46 | 4.19 | — | — | — | — | — | — | — | — | |
| AdaShield-ATarget Model=MiniGPT42025.12 | 7.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OriginalModel=Qwen3-VL-8B2026.02 | 7.47 | 1.03 | 2.06 | 3.09 | 0 | 15.91 | 9.09 | 9.17 | 13.76 | 22.02 | 2.45 | 4 | 4.91 | 11.11 | 15.28 | 13.19 | 1.3 | 1.59 | 4.55 | — | — | — | — | — | — | — | — | |
| MLLM-ProtectorModels=Qwen2.5-VL2026.05 | 8.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| shiftDCModel=DeepSeek-VL-7B2026.02 | 8.48 | 4.02 | 12.31 | 13.92 | 0.68 | 15 | 8.87 | 1.1 | 7.16 | 5.5 | 2.39 | 11.23 | 7.73 | 5.42 | 13.75 | 13.12 | 3.12 | 15 | 12.27 | — | — | — | — | — | — | — | — | |
| HiddenDetectCategory=MLLM Guardrails2025.10 | 8.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdaShield-STarget Model=MiniGPT42025.12 | 9.82 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7BCategory=Offline MLLMs2025.10 | 10 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ECSOModels=Qwen2.5-VL2026.05 | 12.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdashieldModel=DeepSeek-VL-7B2026.02 | 12.72 | 6.03 | 18.47 | 20.88 | 1.02 | 22.5 | 13.3 | 1.65 | 10.73 | 8.26 | 3.59 | 16.84 | 11.6 | 8.13 | 20.62 | 19.69 | 4.68 | 22.5 | 18.41 | — | — | — | — | — | — | — | — | |
| MRDModels=Qwen2.5-VL2026.05 | 12.86 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Claude-3.5-SonnetCategory=Online MLLMs2025.10 | 13.08 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| shiftDCModel=LLaVA-1.6-7B2026.02 | 14.83 | 8.66 | 21.96 | 21.96 | 4.77 | 21.14 | 20.45 | 2.48 | 15.41 | 11.56 | 5.34 | 18.04 | 15.46 | 8.33 | 21.25 | 20.21 | 6.43 | 21.82 | 21.62 | — | — | — | — | — | — | — | — | |
| VanillaModels=Qwen2.5-VL2026.05 | 15.13 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPT-4oCategory=Online MLLMs2025.10 | 16.15 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| JailDAMCategory=MLLM Guardrails2025.10 | 16.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BaseModel=LLaVA-v1.6-Mistral-7B2026.04 | 17.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AM3SafetyBackbone=Qwen2.5-VL-7B2026.01 | 17.26 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 56.01 | 71.23 | — | — | — | — | — | — | |
| Gemma-3-12BInteraction Protocol=Image-tool interaction2026.05 | 21.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdashieldModel=LLaVA-1.6-7B2026.02 | 22.24 | 12.99 | 32.94 | 32.94 | 7.16 | 31.7 | 30.68 | 3.72 | 23.12 | 17.34 | 8.01 | 27.05 | 23.19 | 12.5 | 31.87 | 30.31 | 9.64 | 32.73 | 32.44 | — | — | — | — | — | — | — | — | |
| Qwen3-VL-8BInteraction Protocol=Image-tool interaction2026.05 | 22.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Gemma-3-27BInteraction Protocol=Image-tool interaction2026.05 | 23.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-235BInteraction Protocol=Image-tool interaction2026.05 | 23.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoCAModel=DeepSeek-VL-7B2026.02 | 25.43 | 12.06 | 36.94 | 41.75 | 2.04 | 45 | 26.6 | 3.3 | 21.47 | 16.52 | 7.18 | 33.68 | 23.19 | 16.25 | 41.25 | 39.38 | 9.35 | 45 | 36.82 | — | — | — | — | — | — | — | — | |
| ECSOModel=DeepSeek-VL-7B2026.02 | 26 | 12.33 | 37.76 | 42.68 | 2.09 | 46 | 27.19 | 3.38 | 21.94 | 16.88 | 7.34 | 34.43 | 23.71 | 16.62 | 42.16 | 40.25 | 9.56 | 46 | 37.64 | — | — | — | — | — | — | — | — | |
| ECSOModels=LLaVA-1.5-7b2026.05 | 26.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SPA-VLBackbone=Qwen2.5-VL-7B2026.01 | 27.56 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 44.87 | 53.68 | — | — | — | — | — | — | |
| Qwen3-VL-32BInteraction Protocol=Image-tool interaction2026.05 | 28.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OriginalModel=DeepSeek-VL-7B2026.02 | 28.26 | 13.4 | 41.04 | 46.39 | 2.27 | 50 | 29.55 | 3.67 | 23.85 | 18.35 | 7.98 | 37.42 | 25.77 | 18.06 | 45.83 | 43.75 | 10.39 | 50 | 40.91 | — | — | — | — | — | — | — | — | |
| LLaVA-1.5-7B (base)Category=Offline MLLMs2025.10 | 28.85 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SafeSteerModels=LLaVA-1.5-7b2026.05 | 30.37 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Safe RLHF-VBackbone=Qwen2.5-VL-7B2026.01 | 31.01 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 53.68 | 61.72 | — | — | — | — | — | — | |
| RLHF-VBackbone=Qwen2.5-VL-7B2026.01 | 32.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 4.4 | 9.2 | — | — | — | — | — | — | |
| LlavaGuardCategory=MLLM Guardrails2025.10 | 32.58 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-8BInteraction Protocol=Prior turn (NORMAL)2026.05 | 33.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-8BInteraction Protocol=Direct2026.05 | 34.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Llama-4-ScoutInteraction Protocol=Image-tool interaction2026.05 | 34.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MLLM-ProtectorModels=LLaVA-1.5-7b2026.05 | 35.15 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-235BInteraction Protocol=Prior turn (HARMFUL)2026.05 | 35.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ECSOTarget Model=Llava 22025.12 | 36.01 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-8BInteraction Protocol=Prior turn (HARMFUL)2026.05 | 36.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-235BInteraction Protocol=Direct2026.05 | 36.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VanillaModels=LLaVA-1.5-7b2026.05 | 37.62 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-32BInteraction Protocol=Prior turn (NORMAL)2026.05 | 38.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-235BInteraction Protocol=Prior turn (NORMAL)2026.05 | 38.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SPA-VLBackbone=LLaVA-NeXT-7B2026.01 | 38.21 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 37.71 | 64.14 | — | — | — | — | — | — | |
| MRDModels=LLaVA-1.5-7b2026.05 | 38.31 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pixtral-LargeInteraction Protocol=Image-tool interaction2026.05 | 39.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-32BInteraction Protocol=Prior turn (HARMFUL)2026.05 | 39.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| No DefenseTarget Model=Llava 22025.12 | 39.82 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ECSOTarget Model=MiniGPT42025.12 | 41.52 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Llama-4-ScoutInteraction Protocol=Direct2026.05 | 41.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MM-DPOBackbone=Qwen2.5-VL-7B2026.01 | 41.79 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 47.61 | 53.24 | — | — | — | — | — | — | |
| Gemma-3-27BInteraction Protocol=Direct2026.05 | 42.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-32BInteraction Protocol=Direct2026.05 | 42.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Gemma-3-27BInteraction Protocol=Prior turn (NORMAL)2026.05 | 43.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AM3SafetyBackbone=LLaVA-NeXT-7B2026.01 | 43.33 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 74.05 | 73.2 | — | — | — | — | — | — | |
| Gemma-3-12BInteraction Protocol=Prior turn (NORMAL)2026.05 | 44.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Llama-4-ScoutInteraction Protocol=Prior turn (NORMAL)2026.05 | 44.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RLHF-VBackbone=LLaVA-NeXT-7B2026.01 | 44.17 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 9.25 | 20.05 | — | — | — | — | — | — | |
| CoCAModel=LLaVA-1.6-7B2026.02 | 44.48 | 25.98 | 65.88 | 65.88 | 14.32 | 63.41 | 61.36 | 7.43 | 46.24 | 34.68 | 16.01 | 54.11 | 46.38 | 25 | 63.75 | 60.62 | 19.29 | 65.46 | 64.87 | — | — | — | — | — | — | — | — | |
| Llama-4-ScoutInteraction Protocol=Prior turn (HARMFUL)2026.05 | 44.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Gemma-3-12BInteraction Protocol=Direct2026.05 | 45 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ECSOModel=LLaVA-1.6-7B2026.02 | 45.47 | 26.56 | 67.34 | 67.34 | 14.64 | 64.81 | 62.73 | 7.6 | 47.27 | 35.45 | 16.37 | 55.31 | 47.41 | 25.56 | 65.16 | 61.97 | 19.72 | 66.91 | 66.31 | — | — | — | — | — | — | — | — | |
| No DefenseTarget Model=MiniGPT42025.12 | 45.83 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pixtral-LargeInteraction Protocol=Prior turn (HARMFUL)2026.05 | 46.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Gemma-3-12BInteraction Protocol=Prior turn (HARMFUL)2026.05 | 47.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MM-DPOBackbone=LLaVA-NeXT-7B2026.01 | 48.21 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 55.03 | 59.72 | — | — | — | — | — | — | |
| Gemma-3-27BInteraction Protocol=Prior turn (HARMFUL)2026.05 | 48.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7BAlignment=Original Base Model2026.01 | 48.63 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 50 | 50 | — | — | — | — | — | — | |
| OriginalModel=LLaVA-1.6-7B2026.02 | 49.42 | 28.87 | 73.2 | 73.2 | 15.91 | 70.45 | 68.18 | 8.26 | 51.38 | 38.53 | 17.79 | 60.12 | 51.53 | 27.78 | 70.83 | 67.36 | 21.43 | 72.73 | 72.08 | — | — | — | — | — | — | — | — | |
| Pixtral-LargeInteraction Protocol=Prior turn (NORMAL)2026.05 | 51 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pixtral-LargeInteraction Protocol=Direct2026.05 | 54 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaVA-NeXT-7BAlignment=Original Base Model2026.01 | 61.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 50 | 50 | — | — | — | — | — | — | |
| Safe RLHF-VBackbone=LLaVA-NeXT-7B2026.01 | 62.08 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 45.72 | 47.46 | — | — | — | — | — | — | |
| Llama-Guard3-VisionCategory=MLLM Guardrails2025.10 | 74.89 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Circuit BreakersModel=LLaVA-1.5-7B2026.01 | 97.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TraceRouterModel=LLaVA-1.5-7B2026.01 | 99.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QWEN2.5-VL-7B-INSTRUCTAllowed Domain=Base2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 100 | 100 | 100 | 100 | 100 | 100 | |
| QWEN2.5-VL-7B-INSTRUCT (PALETTE)Allowed Domain=IA2026.05 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 30.2 | 92.4 | 62 | 72.6 | 75 | 81.8 |