Mathematical Visual Question Answering on MathVista
85.6AccuracySeed1.5-VL
Evaluation Results
| Method | Links | |
|---|---|---|
| Seed1.5-VLFrames=-2026.03 | 85.6 | |
| Gemini-2.5-ProFrames=-2026.03 | 82.7 | |
| Gemini 2.5 Pro2026.04 | 82.7 | |
| OpenVLThinkerV2Parameters=8B2026.04 | 79.5 | |
| InternVL3.5-8BFrames=-2026.03 | 78.4 | |
| Qwen3-VL GRPORL Strategy=GRPO2026.04 | 78.1 | |
| Qwen3-VL GDPORL Strategy=GDPO2026.04 | 78 | |
| OneThinker-8BParameters=8B2026.04 | 77.6 | |
| InternVL3.5-4BFrames=-2026.03 | 77.1 | |
| Vision-G12026.04 | 76.1 | |
| VL-Rethinker-7BParameters=7B2026.04 | 74.9 | |
| ARES-7BParameters=7B2026.04 | 74.6 | |
| Qwen3-VL-8B-InstructFrames=642026.03 | 74.2 | |
| Qwen3-VL-Instruct-8BParameters=8B2026.04 | 74.2 | |
| Qwen2.5-VL-72B2025.10 | 74 | |
| Qwen2.5-VL-72B + CARESCost=-7%2025.10 | 74 | |
| VisionZero2026.04 | 73.1 | |
| MM-Eureka-7BParameters=7B2026.04 | 73 | |
| R-MSD (4B)Frames=642026.03 | 72.1 | |
| OVR-7BParameters=7B2026.04 | 72.1 | |
| InternVL3.5-2BFrames=-2026.03 | 71.8 | |
| InternVL3-8BFrames=-2026.03 | 71.6 | |
| Original SFT+RL (4B)Frames=642026.03 | 71.2 | |
| OpenVLThinker-7BParameters=7B2026.04 | 70.2 | |
| Qwen3-VL-4B-InstructFrames=642026.03 | 69.5 | |
| BasePruning Ratio=100%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 69.1 | |
| InternVL3-8B2025.10 | 69 | |
| InternVL3-8B + CARESCost=-22%2025.10 | 69 | |
| InternVL3-8B + CARES-ARCost=-31%2025.10 | 68 | |
| VideoLLaMA3-7BFrames=-2026.03 | 67.1 | |
| InternVL2.5-8BFrames=-2026.03 | 64.4 | |
| GPT-4o2025.10 | 64 | |
| GPT-4oFrames=-2026.03 | 63.8 | |
| GPT-4o2026.04 | 63.8 | |
| Full CacheArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 63.1 | |
| RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.6 | |
| SnapKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 62.2 | |
| MedaArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 61.8 | |
| R-MSD (2B)Frames=642026.03 | 61.4 | |
| LOOK-MArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 61.4 | |
| H2OArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 61.3 | |
| GPT-4o + CARESCost=-17%2025.10 | 61 | |
| InternVL2.5-4BFrames=-2026.03 | 60.5 | |
| SAINTArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 60.5 | |
| Full CacheArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 60.4 | |
| H2OArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 59.3 | |
| RetentiveKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 59.3 | |
| VideoLLaMA3-2BFrames=-2026.03 | 59.2 | |
| MedaArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 59.2 | |
| SnapKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 59.1 | |
| LOOK-MArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 59.1 | |
| Qwen3-VL-2B-InstructFrames=642026.03 | 58.9 | |
| SAINTArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 58.6 | |
| InternVL3-2BFrames=-2026.03 | 57 | |
| ERASEPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 56.4 | |
| ERASEPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 55.1 | |
| iLLaVAPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 54.9 | |
| iLLaVAPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 53.6 | |
| IVC-PrunePruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 52.6 | |
| DARTPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 51.9 | |
| IVC-PrunePruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 51.5 | |
| InternVL2.5-2BFrames=-2026.03 | 51.3 | |
| ERASEPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 50.7 | |
| Full CacheArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 50.4 | |
| PruneSIDPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 50.3 | |
| DARTPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 49.8 | |
| SnapKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 49.5 | |
| iLLaVAPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 49.5 | |
| RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 49.3 | |
| MedaArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 48.8 | |
| LOOK-MArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 48.5 | |
| SAINTArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 48.3 | |
| PruneSIDPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 48.1 | |
| Granite-Vision -2B2025.10 | 48 | |
| Granite-Vision -2B + CARESCost=-22%2025.10 | 48 | |
| H2OArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 48 | |
| IVC-PrunePruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 47.1 | |
| DARTPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 45.1 | |
| Granite-Vision -2B + CARES-ARCost=-25%2025.10 | 44 | |
| PruneSIDPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 43.8 | |
| CDPrunerPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 39.7 | |
| CDPrunerPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 39.6 | |
| SFT Vision2025.10 | 38.4 | |
| MergeMixbaseline=SFT Vision2025.10 | 37.8 | |
| CDPrunerPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 34.1 | |
| VisionThink-7B2025.10 | 23.8 | |
| Qwen2.5-VL-Ins-7B2025.10 | 19.2 |