List your company

Mathematical Visual Question Answering on MathVista

85.6AccuracySeed1.5-VL
16.54434.47252.470.328Oct 22, 2025Nov 24, 2025Dec 28, 2025Jan 30, 2026Mar 5, 2026Apr 7, 2026May 11, 2026

Evaluation results

MethodLinks
Seed1.5-VLFrames=-2026.0385.6
Gemini-2.5-ProFrames=-2026.0382.7
Gemini 2.5 Pro2026.0482.7
OpenVLThinkerV2Parameters=8B2026.0479.5
InternVL3.5-8BFrames=-2026.0378.4
Qwen3-VL GRPORL Strategy=GRPO2026.0478.1
Qwen3-VL GDPORL Strategy=GDPO2026.0478
OneThinker-8BParameters=8B2026.0477.6
InternVL3.5-4BFrames=-2026.0377.1
Vision-G12026.0476.1
VL-Rethinker-7BParameters=7B2026.0474.9
ARES-7BParameters=7B2026.0474.6
Qwen3-VL-8B-InstructFrames=642026.0374.2
Qwen3-VL-Instruct-8BParameters=8B2026.0474.2
Qwen2.5-VL-72B2025.1074
Qwen2.5-VL-72B + CARESCost=-7%2025.1074
VisionZero2026.0473.1
MM-Eureka-7BParameters=7B2026.0473
R-MSD (4B)Frames=642026.0372.1
OVR-7BParameters=7B2026.0472.1
InternVL3.5-2BFrames=-2026.0371.8
InternVL3-8BFrames=-2026.0371.6
Original SFT+RL (4B)Frames=642026.0371.2
OpenVLThinker-7BParameters=7B2026.0470.2
Qwen3-VL-4B-InstructFrames=642026.0369.5
BasePruning Ratio=100%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0569.1
InternVL3-8B2025.1069
InternVL3-8B + CARESCost=-22%2025.1069
InternVL3-8B + CARES-ARCost=-31%2025.1068
VideoLLaMA3-7BFrames=-2026.0367.1
InternVL2.5-8BFrames=-2026.0364.4
GPT-4o2025.1064
GPT-4oFrames=-2026.0363.8
GPT-4o2026.0463.8
Full CacheArchitecture=Qwen3-VL, Parameter Scale=8B2026.0463.1
RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.0462.6
SnapKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.0462.2
MedaArchitecture=Qwen3-VL, Parameter Scale=8B2026.0461.8
R-MSD (2B)Frames=642026.0361.4
LOOK-MArchitecture=Qwen3-VL, Parameter Scale=8B2026.0461.4
H2OArchitecture=Qwen3-VL, Parameter Scale=8B2026.0461.3
GPT-4o + CARESCost=-17%2025.1061
InternVL2.5-4BFrames=-2026.0360.5
SAINTArchitecture=Qwen3-VL, Parameter Scale=8B2026.0460.5
Full CacheArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.0460.4
H2OArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.0459.3
RetentiveKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.0459.3
VideoLLaMA3-2BFrames=-2026.0359.2
MedaArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.0459.2
SnapKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.0459.1
LOOK-MArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.0459.1
Qwen3-VL-2B-InstructFrames=642026.0358.9
SAINTArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.0458.6
InternVL3-2BFrames=-2026.0357
ERASEPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0556.4
ERASEPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0555.1
iLLaVAPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0554.9
iLLaVAPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0553.6
IVC-PrunePruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0552.6
DARTPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0551.9
IVC-PrunePruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0551.5
InternVL2.5-2BFrames=-2026.0351.3
ERASEPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0550.7
Full CacheArchitecture=Qwen3-VL, Parameter Scale=4B2026.0450.4
PruneSIDPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0550.3
DARTPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0549.8
SnapKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.0449.5
iLLaVAPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0549.5
RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.0449.3
MedaArchitecture=Qwen3-VL, Parameter Scale=4B2026.0448.8
LOOK-MArchitecture=Qwen3-VL, Parameter Scale=4B2026.0448.5
SAINTArchitecture=Qwen3-VL, Parameter Scale=4B2026.0448.3
PruneSIDPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0548.1
Granite-Vision -2B2025.1048
Granite-Vision -2B + CARESCost=-22%2025.1048
H2OArchitecture=Qwen3-VL, Parameter Scale=4B2026.0448
IVC-PrunePruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0547.1
DARTPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0545.1
Granite-Vision -2B + CARES-ARCost=-25%2025.1044
PruneSIDPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0543.8
CDPrunerPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0539.7
CDPrunerPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0539.6
SFT Vision2025.1038.4
MergeMixbaseline=SFT Vision2025.1037.8
CDPrunerPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0534.1
VisionThink-7B2025.1023.8
Qwen2.5-VL-Ins-7B2025.1019.2