List your company

Multimodal Understanding on MME (Score)

2,393ScoreInternVL3
1,471.041,710.3951,949.752,189.105Mar 18, 2025Jun 3, 2025Aug 19, 2025Nov 4, 2025Jan 20, 2026Apr 7, 2026Jun 23, 2026

Evaluation results

MethodLinks
InternVL3Params=8B2026.042,393
VisionZipTokens Retained Percentage=33.3%2025.032,387
TwigVLM++Tokens Retained Percentage=33.3%2025.032,373
Qwen3-VLParams=8B2026.042,364
InternVL3.5Params=8B2026.042,357
Qwen2.5-VL-7B2025.122,347
Q2.5VL-7BTokens Retained Percentage=100%2025.032,347
TwigVLM++Tokens Retained Percentage=22.2%2025.032,346
InternVL2.5-8B2025.122,344
Qwen2.5-OmniParams (B)=7, Time (s)=6.02025.082,340
InternVL2.5-4BComplexity=O(n)2025.122,338
TwigVLMTokens Retained Percentage=33.3%2025.032,338
Qwen2-VL-7B2025.122,326
DART + RandomBackbone=Qwen2.5-VL-7B, Token Retention Ratio=50%2025.122,318
VisionZipTokens Retained Percentage=22.2%2025.032,318
Qwen2.5-VLParams=7B2026.042,317
Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B, Token Retention Ratio=100%2025.122,313
GPT-4oTime (s)=1.22025.082,310
DivPrune+VTWBackbone=Qwen2.5-VL-7B, Token Retention Ratio=50%2025.122,308
DART+VTWBackbone=Qwen2.5-VL-7B, Token Retention Ratio=50%2025.122,302
DivPrune + RandomBackbone=Qwen2.5-VL-7B, Token Retention Ratio=50%2025.122,302
FastVTokens Retained Percentage=33.3%2025.032,299
DARTBackbone=Qwen2.5-VL-7B, Token Retention Ratio=50%2025.122,295
DivPruneBackbone=Qwen2.5-VL-7B, Token Retention Ratio=50%2025.122,291
CoT4DET-7B2025.122,289
Qwen-VL-Max2025.082,281
FastVTokens Retained Percentage=22.2%2025.032,246
TwigVLM++Tokens Retained Percentage=11.1%2025.032,240
TwigVLMTokens Retained Percentage=22.2%2025.032,238
InternVL2-8B2025.122,210
Qwen2.5VL-3BComplexity=O(n)2025.122,171
VisionZipTokens Retained Percentage=11.1%2025.032,153
InfiniteVL-4BComplexity=O(1)2025.122,126
TwigVLMTokens Retained Percentage=11.1%2025.032,020
VITA1.5Params (B)=7, Time (s)=3.72025.082,006
LLaVA-OV-7B2025.121,998
Qwen2.5-VLParams (B)=722025.081,980
PAIBackbone=Qwen-VL-Chat2026.051,923
Ours:base Qwen2.5Params (B)=14+72, Time (s)=3.22025.081,922
VDD-NoneBackbone=Qwen-VL-Chat2026.051,921
Qwen2.5-VLParams (B)=322025.081,915
BRACSBackbone=Qwen-VL-Chat2026.051,882
BRACSBackbone=LLaVA-1.5-7B2026.051,880
UniDriveVLAParams=8B2026.041,876
GreedyBackbone=LLaVA-1.5-7B2026.051,866
VanillaToken Budget (K)=576, Token Reduction (%)=100%, Model Backbone=LLaVA-1.5-7B2026.061,864
VanillaEarly Cmp.=−, Retained Tokens=576, Compression Ratio=100%2026.031,862
VCDBackbone=Qwen-VL-Chat2026.051,860
FastVTokens Retained Percentage=11.1%2025.031,859
VCDBackbone=LLaVA-1.5-7B2026.051,853
Phi-3.5-Vision-4BComplexity=O(n)2025.121,846
PriorTRToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.061,845
SPINBackbone=Qwen-VL-Chat2026.051,824
PriorTRToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.061,820
GreedyBackbone=Qwen-VL-Chat2026.051,808
PRUNESID-DynEarly Cmp.=✓, Retained Tokens=192, Compression Ratio=66.7%2026.031,797
VisPrunerToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.061,796
PAIBackbone=LLaVA-1.5-7B2026.051,775
VDD-NoneBackbone=LLaVA-1.5-7B2026.051,772
MaTVLM-3BComplexity=O(n)2025.121,771
SPINBackbone=LLaVA-1.5-7B2026.051,768
PDropToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.061,766
SmolVLM2-2BComplexity=O(n)2025.121,764
VisPrunerToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.061,761
PRUNESID-DynEarly Cmp.=✓, Retained Tokens=128, Compression Ratio=77.8%2026.031,760
PriorTRToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.061,745
PRUNESID-DynEarly Cmp.=✓, Retained Tokens=64, Compression Ratio=88.9%2026.031,734
TinyLLaVA-3BComplexity=O(n)2025.121,733
SparseVLMToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.061,721
SparseVLMToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.061,696
Bagel2026.041,687
BAGALType=Und., # Params=14B2025.031,687
VisPrunerToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.061,687
BLIP3o 8B2026.041,683
Qwen2.5-VLParams (B)=72025.081,673
PaliGemma2-3BComplexity=O(n)2025.121,658
LatentUMBasequantized setting=false2026.041,654
PDropToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.061,644
LatentUMBasequantized setting=true2026.041,638
LLaVA-NeXTType=Und., # Params=34B2025.031,631
PruMergeToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.061,626
DualToken-7BType=Uni., # Params=7B, Resolution=384px2025.031,625
Show-o2Type=Uni., # Params=7B, Resolution=432px2025.031,620.5
Show-o22026.041,620
FastVToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.061,605
DualToken-3BType=Uni., # Params=3B, Resolution=384px2025.031,588.4
ShareGPT4VType=Und., # Params=7B2025.031,567.4
Tokenflow XL2026.041,551
PruMergeToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.061,549
VIFLLM=Vicuna-7B, Resolution=336 × 3362026.041,547.2
TokenFlowType=Uni., # Params=13B, Resolution=384px2025.031,545.9
PruMergeToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.061,544
VILAType=Und., # Params=7B2025.031,533
BLIP3o 4B2026.041,528
LLaVA-HRLLM=Vicuna-7B, Resolution=384 × 3842026.041,522.3
IGVALLM=Vicuna-7B, Resolution=336 × 3362026.041,519.8
LLaVA-NeXTLLM=Mistral-7B, Resolution=672 × 6722026.041,519.3
LLaVA-NeXTType=Und., # Params=7B2025.031,519
LLaVA-1.5Type=Und., # Params=7B2025.031,510.7
Original*Backbone=LLaVA-1.5-7B, Visual Token #=576, TFLOPs (ratio %)=2.23 (100)2026.031,506.5