List your company
ResearchFormerly WizWand

Diagram Understanding on AI2D (Accuracy and Avg. %)

84.5AccuracyQwen2.5-VL-32B
67.23671.71876.280.682Apr 1, 2026Apr 11, 2026Apr 22, 2026May 3, 2026May 14, 2026May 25, 2026Jun 5, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Qwen2.5-VL-32BToken Budget=100%, Backbone Model=Qwen2.5-VL-32B2026.0484.5100
InternVL-2.5Size=8B2026.0684.568
InternVL2.5-MPOSize=8B2026.0684.269.4
MMS-PRMSize=8B2026.0684.273
Qwen2-VLSize=7B2026.068365.1
DivPruneToken Budget=35.3%, Backbone Model=Qwen2.5-VL-32B2026.0482.293.4
IWPToken Budget=35.3%, Backbone Model=Qwen2.5-VL-32B2026.0482.294.2
MiniCPM-V-2-6Size=8B2026.0682.164.2
VisionZipToken Budget=35.3%, Backbone Model=Qwen2.5-VL-32B2026.0481.994
InternVL2.5-MPO + SFTSize=8B2026.0681.671
IXC-2.5Size=7B2026.0681.5—
LLaVA-OneVisionSize=7B2026.0681.464.6
LlamaV-o1Size=11B2026.0681.2—
Insight-VSize=7B2026.0679.865.7
IWPToken Budget=22.2%, Backbone Model=Qwen2.5-VL-32B2026.0479.489.6
VisionZipToken Budget=22.2%, Backbone Model=Qwen2.5-VL-32B2026.0478.986.2
CDPrunerToken Budget=35.3%, Backbone Model=Qwen2.5-VL-32B2026.0478.887.2
LLaVA-CoTSize=11B2026.0678.7—
FastVToken Budget=35.3%, Backbone Model=Qwen2.5-VL-32B2026.0478.690.8
DivPruneToken Budget=22.2%, Backbone Model=Qwen2.5-VL-32B2026.0477.987.7
LLaVA-NeXT-LLaMA3 + SFTSize=8B2026.0676.160.8
FastVToken Budget=22.2%, Backbone Model=Qwen2.5-VL-32B2026.0475.384
IWPToken Budget=11.1%, Backbone Model=Qwen2.5-VL-32B2026.047481.5
VisionZipToken Budget=11.1%, Backbone Model=Qwen2.5-VL-32B2026.047375.1
CDPrunerToken Budget=22.2%, Backbone Model=Qwen2.5-VL-32B2026.0472.981.3
DivPruneToken Budget=11.1%, Backbone Model=Qwen2.5-VL-32B2026.0472.478.9
LLaVA-NeXT-LLaMA3Size=8B2026.0671.552.1
FastVToken Budget=11.1%, Backbone Model=Qwen2.5-VL-32B2026.0470.473.2
CDPrunerToken Budget=11.1%, Backbone Model=Qwen2.5-VL-32B2026.0467.971.8