List your company

Text-based Visual Question Answering on TextVQA

88.69TextVQA AccuracyBF16
56.356464.750773.14581.5393Jun 8, 2025Aug 10, 2025Oct 13, 2025Dec 16, 2025Feb 17, 2026Apr 22, 2026Jun 25, 2026

Evaluation results

MethodLinks
BF16Model=Kimi-VL-A3B-Instruct, Avg-Bit=BF162026.0688.69
BAGEL (Ours)Baseline=BAGEL, Params=7B active2026.0688.5
MODEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.0686.24
BAGELBaseline=BAGEL, Params=7B active2026.0686
BLIP3o-8B (Ours)Baseline=BLIP3o-8B, Params=8B2026.0685.2
MC-MoEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.0685.06
VARGPT-v1.1 (Ours)Baseline=VARGPT-v1.1, Params=7B+2B2026.0684.8
VEQ-MEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.0684.24
BF16Model=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=BF162026.0683.37
BLIP3o-8BBaseline=BLIP3o-8B, Params=8B2026.0683.1
Speed-QModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.0683.01
MoEQuantModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.0682.93
GPTQModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.0682.05
VARGPT-v1.1Baseline=VARGPT-v1.1, Params=7B+2B2026.0682
MBQModel=Kimi-VL-A3B-Instruct, Avg-Bit=W32026.0681.73
VanillaAverage Token Reduction=0.0%, Backbone=Qwen3-VL-8B2026.0681.6
MODEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.0680.65
Qwen2-VL-2B#Params=2B, TFLOPs=0.42025.0679.7
LaTtE-Flow#Params=2B, TFLOPs=0.42025.0679.7
MODEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.0679.38
VEQ-MEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.0678.58
Speed-QModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.0678.25
MBQModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.0677.98
MC-MoEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.0677.93
PDrop+RerouteAverage Token Reduction=66.7%, Backbone=Qwen3-VL-8B2026.0677.8
MoEQuantModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.0677.05
VanillaBackbone=InternVL-2.5-8B, Token Reduction Ratio=0%2026.0676.8
FastVAverage Token Reduction=66.7%, Backbone=Qwen3-VL-8B2026.0676.7
GPTQModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W32026.0676.58
CLSE localBackbone=InternVL-2.5-8B, Token Reduction Ratio=66.7%, FFT Type=Local2026.0675.9
CLSE globalBackbone=InternVL-2.5-8B, Token Reduction Ratio=66.7%, FFT Type=Global2026.0675.9
FastVBackbone=InternVL-2.5-8B, Token Reduction Ratio=66.7%2026.0675.5
MODEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.0675.45
MC-MoEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.0675.15
CLSE globalBackbone=InternVL-2.5-8B, Token Reduction Ratio=77.8%, FFT Type=Global2026.0675
CLSE localBackbone=InternVL-2.5-8B, Token Reduction Ratio=77.8%, FFT Type=Local2026.0674.9
VEQ-MEModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.0674.06
AsymVLMKeep ratio=75%, FLOPs Saved=28%2026.0573.78
FastVBackbone=InternVL-2.5-8B, Token Reduction Ratio=77.8%2026.0673.7
SparseVLMKeep ratio=75%, FLOPs Saved=18%2026.0573.66
BaselineKeep ratio=100%, FLOPs Saved=—2026.0573.54
FastVKeep ratio=75%, FLOPs Saved=0%2026.0573.48
AsymVLMKeep ratio=65%, FLOPs Saved=39%2026.0573.45
SparseVLMKeep ratio=65%, FLOPs Saved=35%2026.0573.34
Uni-DPOModel=Qwen2-VL-2B2025.0673.23
AsymVLMKeep ratio=50%, FLOPs Saved=54%2026.0573.2
SparseVLMKeep ratio=50%, FLOPs Saved=47%2026.0573.2
FastV+RerouteAverage Token Reduction=66.7%, Backbone=Qwen3-VL-8B2026.0673
FastVKeep ratio=65%, FLOPs Saved=0%2026.0572.66
CLSE localBackbone=InternVL-2.5-8B, Token Reduction Ratio=88.9%, FFT Type=Local2026.0672.5
VEQ-MEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.0672.38
FastVKeep ratio=50%, FLOPs Saved=0%2026.0572.24
ILLUMEBaseline=Vicuna-7B, Params=7B2026.0672.1
CLSE globalBackbone=InternVL-2.5-8B, Token Reduction Ratio=88.9%, FFT Type=Global2026.0671.9
PDrop+RerouteAverage Token Reduction=77.8%, Backbone=Qwen3-VL-8B2026.0671.7
PDropAverage Token Reduction=66.7%, Backbone=Qwen3-VL-8B2026.0671
MC-MoEModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.0670.42
MoEQuantModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.0668.84
FastVAverage Token Reduction=77.8%, Backbone=Qwen3-VL-8B2026.0668.5
Qwen2.5-VL-3BKV Size=2562026.0568.2
EMU2 Chat#Params=34B, TFLOPs=5.42025.0666.6
FastVBackbone=InternVL-2.5-8B, Token Reduction Ratio=88.9%2026.0666.4
Cambrian-4B + Gaze attentionKV Size=144+42026.0566.2
Qwen2.5-VL-3B + InfiniPot-VKV Size=1282026.0566.1
Cambrian-4B + Gaze attentionKV Size=288+42026.0566
LLaVA-Next-LLaMA-3-8BKV Size=7842026.0564.7
EMU3#Params=8B, TFLOPs=4.12025.0664.7
Emu3Baseline=Emu32026.0664.7
Cambrian-4B + Gaze attentionKV Size=72+42026.0564.6
Cambrian-4BKV Size=5762026.0564.2
FastV+RerouteAverage Token Reduction=77.8%, Backbone=Qwen3-VL-8B2026.0664.2
LLaVA-Next-Qwen2-7BKV Size=7842026.0563.6
Cambrian-4B + TokenCompre. + Gaze attentionKV Size=54+42026.0562.7
Speed-QModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.0662.63
Cambrian-8B-CLIP-LKV Size=5762026.0562.4
Cambrian-4B + TokenCompre. + Gaze attentionKV Size=36+42026.0562.3
TokenFlowBaseline=TokenFlow2026.0662.3
Speed-QModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.0662.02
MoEQuantModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.0661.32
VanillaBackbone=LLaVA-1.5-13B, Retained Tokens=5762026.0661.2
VILA-UBaseline=VILA-U2026.0660.8
SimPOModel=Qwen2-VL-2B2025.0660.63
PDrop+RerouteAverage Token Reduction=88.9%, Backbone=Qwen3-VL-8B2026.0660.6
MetaMorph#Params=8B, TFLOPs=1.12025.0660.5
MetaMorphBaseline=MetaMorph2026.0660.5
Qwen2.5-VL-3B + InfiniPot-VKV Size=502026.0560.2
Cambrian-4B + TokenCompre.KV Size=1442026.0560.1
MBQModel=Kimi-VL-A3B-Instruct, Avg-Bit=W22026.0659.98
MBQModel=Qwen3-VL-30B-A3B-Instruct, Avg-Bit=W22026.0659.84
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0659.5
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0659.4
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0659.4
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0659.2
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0658.4
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0658.3
VanillaBackbone=LLaVA-1.5-7B, Retained Tokens=576 (Upper Bound)2026.0658.2
LLaVA-OV-7B + HERMESKV Size=1002026.0558
Zoo-PruneBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.0657.9
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0657.7
STSBackbone=LLaVA-1.5-7B, Retained Tokens=1282026.0657.6