List your company
ResearchFormerly WizWand

Multi-modal Evaluation on MME (Score Metric)

2,389.7MME ScoreInternVL3-8B
1,717.1321,891.7412,066.352,240.959Dec 5, 2024Mar 10, 2025Jun 14, 2025Sep 17, 2025Dec 22, 2025Mar 27, 2026Jul 1, 2026
Updated 3mo ago

Evaluation results

MethodLinks
InternVL3-8BBackbone=InternVL3-8B, Retained Tokens=1280, Reduction Ratio=0%2026.062,389.7
Qwen3-VL-4BModel=Qwen3-VL, Parameter Count=4B2026.042,371.74
VanillaBackbone=Qwen2-VL-7B, Flops Ratio Reduction=0%2026.042,307
VanillaToken Reduction=0% (Upper Bound)2026.042,306
DARTBackbone=InternVL3-8B, Retained Tokens=256, Reduction Ratio=80.0%2026.062,280.4
ERABackbone=InternVL3-8B, Retained Tokens=256, Reduction Ratio=80.0%2026.062,251.7
VisionZipBackbone=InternVL3-8B, Retained Tokens=256, Reduction Ratio=80.0%2026.062,192
DivPruneBackbone=InternVL3-8B, Retained Tokens=256, Reduction Ratio=80.0%2026.062,188.7
DARTBackbone=InternVL3-8B, Retained Tokens=128, Reduction Ratio=90.0%2026.062,180.4
Qwen2.5-VL-3BModel=Qwen2.5-VL, Parameter Count=3B2026.042,144.75
InfiniteVL-4BModel=InfiniteVL, Parameter Count=4B, Result Source=InfiniteVL (Tao et al., 2025) paper2026.042,126
HalfVBackbone=Qwen2-VL-7B, Flops Ratio Reduction=88.9%2026.042,124
ERABackbone=InternVL3-8B, Retained Tokens=128, Reduction Ratio=90.0%2026.062,119.2
DASHToken Reduction=75.00%2026.042,109
LLaVA-OV-Pt + IRABackbone=LLaVA-OV, Training stage=IRA2026.072,109
FastVToken Reduction=75.00%2026.042,101
Qwen3VL-2B-SFTTokens=-2026.052,087.55
DARTBackbone=Qwen2-VL-7B, Flops Ratio Reduction=88.9%2026.042,086
InternVL2-4BModel=InternVL2, Parameter Count=4B, Result Source=OpenVLM Leaderboard (OpenCompass, 2025)2026.042,064.6
VisionZipToken Reduction=75.00%2026.042,058
DARTToken Reduction=75.00%2026.042,056
PruMerge+Token Reduction=75.00%2026.042,044
DivPruneBackbone=InternVL3-8B, Retained Tokens=128, Reduction Ratio=90.0%2026.062,043
InternVL2.5-Pt + IRABackbone=InternVL2.5, Training stage=IRA2026.072,038
LLaVA-OV-Pt + SFTBackbone=LLaVA-OV, Training stage=SFT2026.072,030
HoloVBackbone=Qwen2-VL-7B, Flops Ratio Reduction=88.9%2026.042,006
VisionZipBackbone=InternVL3-8B, Retained Tokens=128, Reduction Ratio=90.0%2026.061,987.8
LLaVA-OneVision-7BModel=LLaVA-OneVision, Parameter Count=7B2026.041,987.63
SpB2.0-VL-5BModel=SpB2.0-VL, Parameter Count=5B2026.041,981.99
InternVL2.5-Pt + SFTBackbone=InternVL2.5, Training stage=SFT2026.071,981
InternVL2-Pt + IRABackbone=InternVL2, Training stage=IRA2026.071,959
FastVBackbone=Qwen2-VL-7B, Flops Ratio Reduction=88.9%2026.041,940
DASHToken Reduction=88.89%2026.041,928
DARTToken Reduction=88.89%2026.041,903
DASHToken Reduction=93.75%2026.041,901
EVIANData Selection Strategy=10K subset2026.041,876.89
FastVToken Reduction=88.89%2026.041,866
VanillaBackbone=LLaVA-1.5-7B, Token Reduction Rate=0.0%2025.051,862
VanillaLoc.=-, Retained Tokens=576, Pruning Ratio=0%2026.031,862
MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=192, Token Reduction Rate=66.7%2025.051,860
MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.051,858
DARTBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.051,856
DARTLoc.=L2, Retained Tokens=192, Pruning Ratio=66.7%2026.031,856
VanillaBackbone=LLaVA-NeXT-7B, Token Retention=2880, Reduction Ratio=100%2026.021,851
VanillaBackbone=LLaVA-Next-7B, Token Reduction Rate=0.0%2025.051,851
LLaVA-NeXT-7BBase Model=LLaVA-NeXT-7B, Retained Tokens=2880, Pruning Ratio=100%2026.041,851
MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=128, Token Reduction Rate=77.8%2025.051,845
MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.051,842
ETCTokens=42026.051,840.46
ETCTokens=22026.051,840.24
DARTBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.051,840
DARTLoc.=L2, Retained Tokens=128, Pruning Ratio=77.8%2026.031,840
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.061,839
ETCTokens=12026.051,838.02
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.061,823
HoloVLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.031,820
VisionZipToken Reduction=88.89%2026.041,819
VanillaBase Model=LLaVA 1.5 13B, Tokens Retained=576, Projector Fine-tuning=false2024.121,818
VanillaBackbone=LLaVA-1.5-13B, Retained Tokens=5762026.061,816
ApETBackbone=LLaVA-NeXT-7B, Token Retention=640, Reduction Ratio=77.8%2026.021,815
SCALEData Selection Strategy=10K subset2026.041,814.97
BLIP-2Data Selection Strategy=10K subset2026.041,810.34
MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=64, Token Reduction Rate=88.9%2025.051,806
MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=64, Token Reduction Rate=88.9%2025.051,806
HoloVLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.031,802
PDropLoc.=L8–L24, Retained Tokens=192, Pruning Ratio=66.7%2026.031,797
PruMerge+Token Reduction=88.89%2026.041,794
InternVL2-Pt + SFTBackbone=InternVL2, Training stage=SFT2026.071,792
PRUNESIDLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.031,791
AutoSelectLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.031,791
FlowCutBase Model=LLaVA-NeXT-7B, Retained Tokens=320, Pruning Ratio=88.9%2026.041,791
VisionZip ‡Base Model=LLaVA 1.5 13B, Tokens Retained=192, Projector Fine-tuning=true2024.121,790
VisionZipBackbone=LLaVA-NeXT-7B, Token Retention=640, Reduction Ratio=77.8%2026.021,787
MustDropBackbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.051,787
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.061,786
ApETBackbone=LLaVA-NeXT-7B, Token Retention=320, Reduction Ratio=88.9%2026.021,783
VisionZipLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.031,783
PDropBackbone=LLaVA-NeXT-7B, Token Retention=640, Reduction Ratio=77.8%2026.021,781
DeSAPBase Model=LLaVA-NeXT-7B, Retained Tokens=320, Pruning Ratio=88.9%2026.041,781
DARTToken Reduction=93.75%2026.041,774
SparseVLMBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.061,774
SparseVLMBackbone=LLaVA-NeXT-7B, Token Retention=640, Reduction Ratio=77.8%2026.021,773
MaTVLM-3BModel=MaTVLM, Parameter Count=3B, Result Source=InfiniteVL (Tao et al., 2025) paper2026.041,771
SmolVLM2-2.2BModel=SmolVLM2, Parameter Count=2.2B2026.041,770.07
SparseVLMBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.061,768
DARTBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=64, Token Reduction Rate=88.9%2025.051,765
AutoSelectLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.031,765
DARTLoc.=L2, Retained Tokens=64, Pruning Ratio=88.9%2026.031,765
VisionZipLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.031,762
PDropLoc.=L8–L24, Retained Tokens=128, Pruning Ratio=77.8%2026.031,761
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=642026.061,761
MoB (with η-prior)Backbone=LLaVA-Next-7B, Objectives=PA VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.051,760
VisionZipBase Model=LLaVA 1.5 13B, Tokens Retained=192, Projector Fine-tuning=false2024.121,754
DivPruneLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.031,752
DivPruneLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.031,752
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.061,751
PRUNESIDLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.031,749
MustDropBackbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.051,745
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.061,744
VisionZipBase Model=LLaVA 1.5 13B, Tokens Retained=128, Projector Fine-tuning=false2024.121,743