List your company
ResearchFormerly WizWand

Multi-modal Understanding on MME (MME score)

2,371.9MME ScoreInternVL3.5-8B
1,681.4441,860.6972,039.952,219.203Apr 23, 2026Apr 27, 2026May 2, 2026May 6, 2026May 11, 2026May 15, 2026May 20, 2026
Updated 5mo ago

Evaluation results

MethodLinks
InternVL3.5-8BRatio=100%, Backbone=InternVL3.5-8B2026.052,371.9
VanillaVisual Tokens=866, Prefill Time (s)=0.2215, FLOPs (T)=2.1015, GPU Mem. (GB)=16.01, SpeedUp=1.00×2026.052,329
DivPruneRatio=60%, Backbone=InternVL3.5-8B2026.052,298.3
F3ARatio=60%, Backbone=InternVL3.5-8B2026.052,291.7
BaselineArchitecture=Qwen-2.5-VL2026.042,279
CDPrunerRatio=60%, Backbone=InternVL3.5-8B2026.052,276.1
Latent DenoisingArchitecture=Qwen-2.5-VL2026.042,265
FastVRatio=60%, Backbone=InternVL3.5-8B2026.052,264.9
VisionZipRatio=60%, Backbone=InternVL3.5-8B2026.052,260
SPprunerVisual Tokens=192, Prefill Time (s)=0.0876, FLOPs (T)=0.5798, GPU Mem. (GB)=15.86, SpeedUp=2.53×2026.052,182
DARTVisual Tokens=192, Prefill Time (s)=0.1068, FLOPs (T)=0.5798, GPU Mem. (GB)=15.86, SpeedUp=2.07×2026.052,174
FastVRatio=40%, Backbone=InternVL3.5-8B2026.052,167.9
CDPrunerRatio=40%, Backbone=InternVL3.5-8B2026.052,164.9
F3ARatio=40%, Backbone=InternVL3.5-8B2026.052,135.7
VisionZipRatio=40%, Backbone=InternVL3.5-8B2026.052,131.4
DivPruneRatio=40%, Backbone=InternVL3.5-8B2026.052,122.4
F3ARatio=20%, Backbone=InternVL3.5-8B2026.052,115.7
CDPrunerRatio=20%, Backbone=InternVL3.5-8B2026.052,101.5
DivPruneRatio=20%, Backbone=InternVL3.5-8B2026.052,099.1
VisionZipRatio=20%, Backbone=InternVL3.5-8B2026.052,051.7
FastVRatio=20%, Backbone=InternVL3.5-8B2026.051,968.7
BaselineArchitecture=LLaVA+CLIP2026.041,879
Latent DenoisingArchitecture=LLaVA+CLIP2026.041,846
Latent DenoisingArchitecture=LLaVA+SigLIP2026.041,733
BaselineArchitecture=LLaVA+SigLIP2026.041,708