List your company
ResearchFormerly WizWand

Vision Understanding on MMBench

85AccuracyBAGEL
60.76867.05973.3579.641Dec 28, 2023May 10, 2024Sep 21, 2024Feb 2, 2025Jun 16, 2025Oct 28, 2025Mar 12, 2026
Updated 3mo ago

Evaluation results

MethodLinks
BAGEL2025.0685——
BLIP3-o2025.0683.5——
MetaQuery2025.0683.5——
UniWorld-V12025.0683.5——
UniMRGType=Unified, # LLM Params=2B2026.0181.44——
InternVL3Type=Und. Only, # LLM Params=2B2026.0181.19——
OpenUniType=Unified, # LLM Params=2B2026.0181.19——
UniLIPType=Unified, # LLM Params=2B2026.0180.7——
BAGELType=Unified, # LLM Params=3B2026.0179.2——
Qwen2.5-VLType=Und. Only, # LLM Params=3B2026.0178.26——
Qwen3-VLType=Und. Only, # LLM Params=2B2026.0177.32——
Janus-Pro2025.0675.5——
ILLUMELLM=Vicuna-7B, Resolution=-2026.0375.1——
DualTokenLLM=LLaMA-2-7B, Resolution=2562026.0374.9——
EvoTokLLM=Qwen-2.5-7B-Inst., Resolution=2562026.0374.9——
UIO-2XXLModel Size=6.8B2023.1271.5——
Unified-IO 2LLM=6.8B from scratch, Resolution=3842026.0371.5——
Janus2025.0669.4——
SIMABase Model=VILA-7B2024.0569.2——
GT-DPOBase Model=VILA-7B2024.0569——
VILA-7BBase Model=VILA-7B2024.0568.9——
VILALLM=LLaMA-2-7B, Resolution=3362026.0368.9——
ShareGPT4VLLM=Vicuna-7B, Resolution=3362026.0368.8——
SIMABase Model=LLaVA-1.5-13B2024.0568.4——
Dynamic-LLaVA-13B^VFree=false, Image (prefill) Token=115, TFLOPs=4.72024.1268.3——
UIO-2XLModel Size=3.2B2023.1268.1——
GT-DPOBase Model=LLaVA-1.5-13B2024.0568——
LLaVA-TokenPacker-13B-144TokenProjector=TokenPacker, Image Resolution=336, Prefill Tokens=144, Prefill TFLOPs=4.92024.1268——
VIG training + LACINGBase Model=LLaVA-1.5 7B2026.0267.89——
LLaVa-1.5Model Size=13B2023.1267.7——
LLaVA-1.5-13BBase Model=LLaVA-1.5-13B2024.0567.7——
LLaVA-1.5-13BImage (prefill) Token=576, TFLOPs=19.62024.1267.7——
LLaVA-v1.5LLM=Vicuna-1.5-13B, Resolution=3362026.0367.7——
VIG training + VARBase Model=LLaVA-1.5 7B2026.0267.66——
VIG training + VCDBase Model=LLaVA-1.5 7B2026.0267.65——
TokLIPLLM=Qwen2.5-7B-Inst., Resolution=3842026.0367.6——
Dynamic-LLaVA-TokenPacker-13B_VProjector=TokenPacker, Image Resolution=336, Prefill Tokens=57, Prefill TFLOPs=2.12024.1267.4——
VanillaBackbone=LLaVA-Next-7B, Token Reduction Rate=0.0%2025.0567.4——
VIG training + PAIBase Model=LLaVA-1.5 7B2026.0267.34——
LLaVA-FastV (13B)Image (prefill) Token=144, TFLOPs=62024.1266.9——
Dynamic-LLaVA-13B^ITFree=false, Image (prefill) Token=115, TFLOPs=4.72024.1266.9——
LLaVA-TokenPacker-FastV-13B k=3, r=0.5Projector=TokenPacker, Image Resolution=336, Prefill Tokens=72, Prefill TFLOPs=2.62024.1266.8——
Dynamic-LLaVA-TokenPacker-13B_VTProjector=TokenPacker, Image Resolution=336, Prefill Tokens=57, Prefill TFLOPs=2.12024.1266.8——
Janus-ProType=Unified, # LLM Params=7B2026.0166.67——
LACINGBase Model=LLaVA-1.5 7B2026.0266.45——
LLaVA-HiRED (7B)Free=true, Image (prefill) Token=115, TFLOPs=22024.1266.4——
VIG trainingBase Model=LLaVA-1.5 7B2026.0266.33——
VARBase Model=LLaVA-1.5 7B2026.0266.3——
LLaVA-LDP-v2-7BProjector=LDP-v2, Image Resolution=336, Prefill Tokens=144, Prefill TFLOPs=2.52024.1266.2——
LLaVA-TokenPacker-13B-64TokenProjector=TokenPacker, Image Resolution=336, Prefill Tokens=64, Prefill TFLOPs=2.22024.1266.2——
MoB (with η-prior)Backbone=LLaVA-Next-7B, Objectives=PA VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.0565.8——
LLaVA-PruMerge+ (13B)Image (prefill) Token=146, TFLOPs=4.92024.1265.7——
LLaVA-1.5 7B2026.0265.46——
PAIBase Model=LLaVA-1.5 7B2026.0265.45——
Dynamic-LLaVA-7B_VProjector=MLP, Image Resolution=336, Prefill Tokens=115, Prefill TFLOPs=2.52024.1265.4——
Dynamic-LLaVA-7B^VFree=false, Image (prefill) Token=115, TFLOPs=2.52024.1265.4——
DARTBackbone=LLaVA-Next-7B, Objectives=VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.0565.3——
LLaVA-TokenPacker-7B-144TokenProjector=TokenPacker, Image Resolution=336, Prefill Tokens=144, Prefill TFLOPs=2.52024.1265.1——
Dynamic-LLaVA-TokenPacker-7B_VProjector=TokenPacker, Image Resolution=336, Prefill Tokens=57, Prefill TFLOPs=1.12024.1265.1——
LLaVA-TokenPacker-FastV-7B k=3, r=0.5Projector=TokenPacker, Image Resolution=336, Prefill Tokens=72, Prefill TFLOPs=1.42024.1265——
SIMABase Model=LLaVA-1.5-7B2024.0564.9——
LLaVA-PruMerge+ (7B)Image (prefill) Token=146, TFLOPs=2.52024.1264.9——
VanillaBase Model=LLaVA-1.5-7B, Retained Tokens=576, Reduction Rate=0%2026.0264.7——
VanillaBackbone=LLaVA-1.5-7B, Token Reduction Rate=0.0%2025.0564.7——
GT-DPOBase Model=LLaVA-1.5-7B2024.0564.6——
mPLUG-Owl2Model Size=8.2B2023.1264.5——
Emu3-ChatType=Und. Only, # LLM Params=8B2026.0164.35——
VCDBase Model=LLaVA-1.5 7B2026.0264.34——
LLaVa-1.5Model Size=7.2B2023.1264.3——
LLaVA-1.5-7BBase Model=LLaVA-1.5-7B2024.0564.3——
LLaVA-1.5-7BProjector=MLP, Image Resolution=336, Prefill Tokens=576, Prefill TFLOPs=10.12024.1264.3——
LLaVA-1.5-7BImage (prefill) Token=576, TFLOPs=10.12024.1264.3——
POVIDBase Model=LLaVA-1.5-7B2024.0564.1——
Dynamic-LLaVA-7B_VTProjector=MLP, Image Resolution=336, Prefill Tokens=115, Prefill TFLOPs=2.52024.1264.1——
LLaVA-TokenPacker-7B-64TokenProjector=TokenPacker, Image Resolution=336, Prefill Tokens=64, Prefill TFLOPs=1.12024.1264.1——
Dynamic-LLaVA-7B^ITFree=false, Image (prefill) Token=115, TFLOPs=2.52024.1264.1——
MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=192, Token Reduction Rate=66.7%2025.0564.1——
LLaVA-Pixel-Shuffle-7BProjector=Pixel-Shuffle, Image Resolution=336, Prefill Tokens=144, Prefill TFLOPs=2.52024.1264——
HA-DPOBase Model=LLaVA-1.5-7B2024.0563.9——
MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.0563.8——
DARTBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.0563.6——
Dynamic-LLaVA-TokenPacker-7B_VTProjector=TokenPacker, Image Resolution=336, Prefill Tokens=57, Prefill TFLOPs=1.12024.1263.5——
LLaVA-FastV (7B)Free=true, Image (prefill) Token=144, TFLOPs=3.22024.1263.5——
MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=128, Token Reduction Rate=77.8%2025.0563.5——
ApETBase Model=LLaVA-1.5-7B, Retained Tokens=192, Reduction Rate=66.7%2026.0263.4——
PDropBase Model=LLaVA-1.5-7B, Retained Tokens=192, Reduction Rate=66.7%2026.0263.3——
DARTBackbone=LLaVA-1.5-7B, Objectives=VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.0563.2——
MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=128, Token Reduction Rate=77.8%2025.0563.2——
LLaVA-Resampler-7BProjector=Resampler, Image Resolution=336, Prefill Tokens=144, Prefill TFLOPs=2.52024.1263.1——
LLaVA-C-Abstractor-7BProjector=C-Abstractor, Image Resolution=336, Prefill Tokens=144, Prefill TFLOPs=2.52024.1263.1——
VisionZipBase Model=LLaVA-1.5-7B, Retained Tokens=192, Reduction Rate=66.7%2026.0263——
MustDropBackbone=LLaVA-Next-7B, Objectives=PA VP, Pruning budget K=320, Token Reduction Rate=88.9%2025.0562.8——
SparseVLMBase Model=LLaVA-1.5-7B, Retained Tokens=192, Reduction Rate=66.7%2026.0262.5——
SparseVLMBackbone=LLaVA-1.5-7B, Objectives=PA, Pruning budget K=192, Token Reduction Rate=66.7%2025.0562.5——
ApETBase Model=LLaVA-1.5-7B, Retained Tokens=128, Reduction Rate=77.8%2026.0262.3——
MustDropBackbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=192, Token Reduction Rate=66.7%2025.0562.3——
UIO-2LModel Size=1.1B2023.1262.1——
MoB (+ η-prior)Backbone=LLaVA-1.5-7B, Objectives=-, Pruning budget K=64, Token Reduction Rate=88.9%2025.0562.1——
VisionZipBase Model=LLaVA-1.5-7B, Retained Tokens=128, Reduction Rate=77.8%2026.0262——
MoB (w/o η-prior)Backbone=LLaVA-1.5-7B, Objectives=PA VP, Pruning budget K=64, Token Reduction Rate=88.9%2025.0561.7——