Multimodal Evaluation on MMStar
69.5AccuracyInternVL2.5-78B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| InternVL2.5-78B2025.06 | 69.5 | — | — | |
| Qwen2-VL-72B2025.06 | 68.3 | — | — | |
| VanillaBackbone=InternVL3.5-8B2026.05 | 68.3 | — | — | |
| VanillaBackbone=InternVL3.5-8B2026.05 | 68.3 | — | — | |
| InternVL2.5-8B-GenRecalTeacher VLM=InternVL2.5-78B2025.06 | 67.8 | — | — | |
| EVA(text)Backbone=InternVL3.5-8B2026.05 | 67.6 | — | — | |
| EVA(text)Backbone=InternVL3.5-8B2026.05 | 67.6 | — | — | |
| EVABackbone=InternVL3.5-8B2026.05 | 67.5 | — | — | |
| EVABackbone=InternVL3.5-8B2026.05 | 67.5 | — | — | |
| InternVL2-76B2025.06 | 67.4 | — | — | |
| VLGuardBackbone=InternVL3.5-8B2026.05 | 67 | — | — | |
| VLGuardBackbone=InternVL3.5-8B2026.05 | 67 | — | — | |
| SophiaVL-R1-7BParameters=7B2025.05 | 66.7 | — | — | |
| JailGuardBackbone=InternVL3.5-8B2026.05 | 66.5 | — | — | |
| JailGuardBackbone=InternVL3.5-8B2026.05 | 66.5 | — | — | |
| InternVL2.5-8B-GenRecalTeacher VLM=Qwen2-VL-72B2025.06 | 66.1 | — | — | |
| LLaVA-OneVision-72BParameters=72B2025.05 | 66.1 | — | — | |
| Jigsaw + CAREModel Size=7B2025.12 | 65.8 | — | — | |
| LLaVA-OneVision-72B2025.06 | 65.8 | — | — | |
| InternVL2.5-8B-GenRecalTeacher VLM=InternVL2-76B2025.06 | 65.8 | — | — | |
| Qwen2.5-VL-7B-InstructParameters=7B, Training Strategy=GRPO2025.05 | 65.6 | — | — | |
| Mix + CL + CAREModel Size=7B2025.12 | 65.2 | — | — | |
| Claude-3.5-Sonnet2025.06 | 65.1 | — | — | |
| InternVL2.5-8B-GenRecalTeacher VLM=NVLM-72B2025.06 | 65.1 | — | — | |
| GPT-4o (0513)2025.06 | 64.7 | — | — | |
| Qwen2.5-VL-7B-InstructParameters=7B, Training Strategy=SFT+GRPO2025.05 | 64.7 | — | — | |
| Qwen 2.5 VLModel Size=7B2025.12 | 64.67 | — | — | |
| Rotation + CL + CAREModel Size=7B2025.12 | 64.6 | — | — | |
| JigsawModel Size=7B2025.12 | 64.47 | — | — | |
| Qwen2.5-VL-7B-InstructParameters=7B, Training Strategy=Direct2025.05 | 64.3 | — | — | |
| GRPO-CAREModel Size=7B2025.12 | 64.13 | — | — | |
| SAIL-VL2Number of Parameters=2.7B2025.12 | 64.1 | — | — | |
| NVLM-72B2025.06 | 63.7 | — | — | |
| Jigsaw + CL + CAREModel Size=7B2025.12 | 63.6 | — | — | |
| Vision-ZeroModel Size=7B2025.12 | 63.47 | — | — | |
| InternVL2.5-8B-VisualPRMParameters=8B2025.05 | 63.4 | — | — | |
| Molmo-72B2025.06 | 63.3 | — | — | |
| Jigsaw + CLModel Size=7B2025.12 | 63.27 | — | — | |
| VisualSphinxModel Size=7B2025.12 | 63.2 | — | — | |
| SAIL-VL1.5Number of Parameters=2.5B2025.12 | 62.8 | — | — | |
| Full CacheArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.73 | — | — | |
| InternVL3.5Number of Parameters=2.3B2025.12 | 62.7 | — | — | |
| RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.58 | — | — | |
| Visual JigsawModel Size=7B2025.12 | 62.53 | — | — | |
| Full CacheArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 62.52 | — | — | |
| RetentiveKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 62.42 | — | — | |
| MedaArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.41 | — | — | |
| SAINTArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.38 | — | — | |
| SnapKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.36 | — | — | |
| H2OArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.33 | — | — | |
| LOOK-MArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.33 | — | — | |
| ViCritModel Size=7B2025.12 | 62.27 | — | — | |
| LOOK-MArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 62.25 | — | — | |
| SnapKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 62.13 | — | — | |
| MedaArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 62.07 | — | — | |
| InternVL2-8BModel=InternVL2-8B, Pruning Method=None2025.09 | 62 | — | — | |
| BaselineCompression Ratio=0%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 61.88 | — | — | |
| SAINTArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 61.84 | — | — | |
| LLaVA-OneVision-7BParameters=7B2025.05 | 61.7 | — | — | |
| H2OArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 61.66 | — | — | |
| VLGuardBackbone=Qwen2.5-VL-7B2026.05 | 61.6 | — | — | |
| VLGuardBackbone=Qwen2.5-VL-7B2026.05 | 61.6 | — | — | |
| PTPModel=InternVL2-8B, Pruning Method=PTP, Pruning Ratio (r)=0.5, Balancing Weight (alpha)=0.52025.09 | 61.4 | — | — | |
| HyperVL ViTLNumber of Parameters=2.0B2025.12 | 61.1 | — | — | |
| EVA(text)Backbone=Qwen2.5-VL-7B2026.05 | 61.1 | — | — | |
| EVA(text)Backbone=Qwen2.5-VL-7B2026.05 | 61.1 | — | — | |
| EVABackbone=Qwen2.5-VL-7B2026.05 | 60.9 | — | — | |
| EVABackbone=Qwen2.5-VL-7B2026.05 | 60.9 | — | — | |
| InternVL3Number of Parameters=2.1B2025.12 | 60.7 | — | — | |
| Full CacheArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 60.66 | — | — | |
| RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 60.58 | — | — | |
| SnapKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 60.41 | — | — | |
| MedaArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 60.38 | — | — | |
| H2OArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 60.36 | — | — | |
| LOOK-MArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 60.25 | — | — | |
| SAINTArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 60.21 | — | — | |
| VanillaBackbone=Qwen2.5-VL-7B2026.05 | 60.1 | — | — | |
| VanillaBackbone=Qwen2.5-VL-7B2026.05 | 60.1 | — | — | |
| AndesVLNumber of Parameters=2.4B2025.12 | 60 | — | — | |
| PatchFit + CL + CAREModel Size=7B2025.12 | 59.87 | — | — | |
| HyperVLNumber of Parameters=1.8B2025.12 | 59.2 | — | — | |
| JailGuardBackbone=Qwen2.5-VL-7B2026.05 | 59.2 | — | — | |
| JailGuardBackbone=Qwen2.5-VL-7B2026.05 | 59.2 | — | — | |
| Gemini-1.5-Pro2025.06 | 59.1 | — | — | |
| R1-Onevision-7BParameters=7B2025.05 | 59.1 | — | — | |
| VanillaBackbone=InternVL-2.5-8B, Token Reduction Rate=100%2026.04 | 58.4 | — | — | |
| Qwen3-VLNumber of Parameters=2.1B2025.12 | 58.3 | — | — | |
| MoD-Single RoundBackbone=Qwen2.5VL-3b-Instruct, Protocol=single-round inference2026.06 | 57.84 | — | — | |
| MoD-Multi RoundBackbone=Qwen2.5VL-3b-Instruct, Protocol=multi-turn dialectical reasoning2026.06 | 57.47 | — | — | |
| GPT-4V2025.05 | 57.1 | — | — | |
| SCOPECompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 56.86 | — | — | |
| MoE-LoRABackbone=Qwen2.5VL-3b-Instruct2026.06 | 56.85 | — | — | |
| Multi-Agent DebateBackbone=Qwen2.5VL-3b-Instruct2026.06 | 56.67 | — | — | |
| Self-CorrectionBackbone=Qwen2.5VL-3b-Instruct2026.06 | 56.57 | — | — | |
| IDPrunerCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 56.49 | — | — | |
| Qwen2.5VL-3b-Instruct2026.06 | 56.3 | — | — | |
| VisionSelectorCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 55.78 | — | — | |
| VisionZipCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 54.86 | — | — | |
| Ovis2Number of Parameters=2.5B2025.12 | 54.7 | — | — | |
| HiPruneCompression Ratio=75%, Backbone=Qwen-2.5-7B-Instruct2026.02 | 54.43 | — | — |