List your company
ResearchFormerly WizWand

Optical Character Recognition Evaluation on OCRBench

831ScoreFP16
-32.512191.669415.85640.031Oct 16, 2025Nov 21, 2025Dec 27, 2025Feb 1, 2026Mar 9, 2026Apr 14, 2026May 20, 2026
Updated 5mo ago

Evaluation results

MethodLinks
FP16Spiking=false, Time Step=N/A, Backbone=Qwen2VL-72B2026.04831——
VanillaToken Retention Ratio=100%2026.05822——
SpikeMLLM (QuaRot+MSTS+TC-LIF)Spiking=true, Time Step=3/4, Backbone=Qwen2VL-72B2026.04817——
SPprunerToken Retention Ratio=70%2026.05814——
DARTToken Retention Ratio=70%2026.05805——
SPprunerToken Retention Ratio=50%2026.05798——
SpikeMLLM(MBQ)Spiking=true, Time Step=255, Backbone=Qwen2VL-72B2026.04791——
DARTToken Retention Ratio=50%2026.05751——
FastVToken Retention Ratio=70%2026.05711——
SPprunerToken Retention Ratio=30%2026.05704——
SpikeMLLM(SQ)Spiking=true, Time Step=255, Backbone=Qwen2VL-72B2026.04678——
FastVToken Retention Ratio=50%2026.05632——
DARTToken Retention Ratio=30%2026.05628——
FastVToken Retention Ratio=30%2026.05510——
VanillaBackbone=LLaVA-1.5-13B, Tokens=322026.013311002,394
VisionTrimBackbone=LLaVA-1.5-13B, Tokens=322026.0129796.3704
VisionZipBackbone=LLaVA-1.5-13B, Tokens=322026.0116571.41,096
Ovis-U1#LLM=1.5B2025.1288.3——
InternVL3Input=Tile-wise, RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=8B2025.1088——
Qwen2-VLInput=Any Res., RoPE=M-RoPE, Backbone=Dense, Parameter Scale=8B2025.1086.6——
STAR-7B#LLM=7B2025.1286.4——
Qwen2.5-VLInput=Any Res., RoPE=M-RoPE, Backbone=Dense, Parameter Scale=8B2025.1086.4——
Encoder-BasedInput=Tile-wise, RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=8B2025.1085.3——
Qwen3-VLparameters=8B2026.0184.9——
InternVL3Input=Tile-wise, RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=2B2025.1083.5——
Encoder-BasedInput=Tile-wise, RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=2B2025.1083.5——
LLaVA-OVversion=1.5, parameters=8B2026.0182.5——
InternVL2.5Input=Tile-wise, RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=8B2025.1082.2——
MiniCPM-Vversion=4.5, parameters=8B2026.0181.5——
Qwen2-VLInput=Any Res., RoPE=M-RoPE, Backbone=Dense, Parameter Scale=2B2025.1080.9——
InternVL2.5Input=Tile-wise, RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=2B2025.1080.4——
Intern-S1variant=mini, parameters=9B2026.0180.2——
Mono-InternVL-1.5Input=Tile-wise, RoPE=1D-RoPE, Backbone=DaC, Parameter Scale=2B2025.1080.1——
Innovator-VLvariant=8B-Instruct, parameters=8B2026.0180——
InternVL3.5parameters=8B2026.0180——
STAR-3B#LLM=3B2025.1279.7——
Qwen2.5-VLInput=Any Res., RoPE=M-RoPE, Backbone=Dense, Parameter Scale=2B2025.1079.7——
Innovator-VLvariant=8B-Thinking, parameters=8B2026.0179.3——
SAILInput=Any Res., RoPE=M-RoPE, Backbone=Dense, Parameter Scale=8B2025.1078.3——
MiMo-VLtraining=7B-RL, parameters=7B2026.0178——
NEOInput=Any Res., RoPE=Native-RoPE, Backbone=Dense, Parameter Scale=8B2025.1077.7——
NEOInput=Any Res., RoPE=Native-RoPE, Backbone=Dense, Parameter Scale=2B2025.1077.1——
MiMo-VLtraining=7B-SFT, parameters=7B2026.0176.9——
Mono-InternVLInput=Tile-wise, RoPE=1D-RoPE, Backbone=MoE, Parameter Scale=2B2025.1076.7——
HoVLEInput=Tile-wise, RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=2B2025.1074——
X-Omni#LLM=7B2025.1270.4——
EVEv2Input=Any Rat., RoPE=1D-RoPE, Backbone=DaC, Parameter Scale=8B2025.1070.2——
EMU3#LLM=8B2025.1268.7——
Emu3Input=Fix Res., RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=8B2025.1068.7——
ILLUME+#LLM=3B2025.1267.2——
Qwen3-VL-8B-ThinkingStrategy=LongCoT2026.0265.9——
SAPStrategy=SAP2026.0260.2——
Qwen3-VL-8B-InstructStrategy=Instruct2026.0256.8——
EVEInput=Any Rat., RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=8B2025.1039.8——
FuyuInput=Any Res., RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=8B2025.1036.6——
GRIP+decoder+rlBackbone=LLaVA-1.5-7B, Token Budget=192 tokens, Optimization Strategy=rl, Pruning Component=decoder2026.0531.7——
GRIP+decoder+slBackbone=LLaVA-1.5-7B, Token Budget=192 tokens, Optimization Strategy=sl, Pruning Component=decoder2026.0531.6——
OriginalBackbone=LLaVA-1.5-7B, Token Budget=Full2026.0531.3——
VisionZipBackbone=LLaVA-1.5-7B, Token Budget=192 tokens2026.0531.2——
GRIP+encoder+slBackbone=LLaVA-1.5-7B, Token Budget=192 tokens, Optimization Strategy=sl, Pruning Component=encoder2026.0530.9——
SparseVLMBackbone=LLaVA-1.5-7B, Token Budget=192 tokens2026.0530.7——
VisprunerBackbone=LLaVA-1.5-7B, Token Budget=192 tokens2026.0530.6——
GRIP+encoder+rlBackbone=LLaVA-1.5-7B, Token Budget=192 tokens, Optimization Strategy=rl, Pruning Component=encoder2026.0530.6——
DivPruneBackbone=LLaVA-1.5-7B, Token Budget=192 tokens2026.0530.4——
DARTBackbone=LLaVA-1.5-7B, Token Budget=192 tokens2026.0530——
VisionZipBackbone=LLaVA-1.5-7B, Token Budget=128 tokens2026.0529.8——
GRIP+encoder+slBackbone=LLaVA-1.5-7B, Token Budget=128 tokens, Optimization Strategy=sl, Pruning Component=encoder2026.0529.8——
GRIP+encoder+rlBackbone=LLaVA-1.5-7B, Token Budget=128 tokens, Optimization Strategy=rl, Pruning Component=encoder2026.0529.5——
VisprunerBackbone=LLaVA-1.5-7B, Token Budget=128 tokens2026.0529——
DivPruneBackbone=LLaVA-1.5-7B, Token Budget=128 tokens2026.0528.8——
GRIP+encoder+slBackbone=LLaVA-1.5-7B, Token Budget=64 tokens, Optimization Strategy=sl, Pruning Component=encoder2026.0528.6——
PruneSIDBackbone=LLaVA-1.5-7B, Token Budget=192 tokens2026.0528.3——
GRIP+decoder+slBackbone=LLaVA-1.5-7B, Token Budget=128 tokens, Optimization Strategy=sl, Pruning Component=decoder2026.0528.2——
VisionZipBackbone=LLaVA-1.5-7B, Token Budget=64 tokens2026.0528.2——
SparseVLMBackbone=LLaVA-1.5-7B, Token Budget=128 tokens2026.0528.1——
GRIP+decoder+rlBackbone=LLaVA-1.5-7B, Token Budget=128 tokens, Optimization Strategy=rl, Pruning Component=decoder2026.0528.1——
GRIP+encoder+rlBackbone=LLaVA-1.5-7B, Token Budget=64 tokens, Optimization Strategy=rl, Pruning Component=encoder2026.0528——
VisprunerBackbone=LLaVA-1.5-7B, Token Budget=64 tokens2026.0527.9——
DARTBackbone=LLaVA-1.5-7B, Token Budget=128 tokens2026.0527.5——
PruneSIDBackbone=LLaVA-1.5-7B, Token Budget=128 tokens2026.0527.5——
PruneSIDBackbone=LLaVA-1.5-7B, Token Budget=64 tokens2026.0527.4——
DivPruneBackbone=LLaVA-1.5-7B, Token Budget=64 tokens2026.0527.1——
GRIP+decoder+rlBackbone=LLaVA-1.5-7B, Token Budget=64 tokens, Optimization Strategy=rl, Pruning Component=decoder2026.0522.8——
DARTBackbone=LLaVA-1.5-7B, Token Budget=64 tokens2026.0521.4——
SparseVLMBackbone=LLaVA-1.5-7B, Token Budget=64 tokens2026.0521.2——
GRIP+decoder+slBackbone=LLaVA-1.5-7B, Token Budget=64 tokens, Optimization Strategy=sl, Pruning Component=decoder2026.0521.2——
SOLOInput=Any Res., RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=8B2025.1012.6——
DeepEyes2025.120.846——
SubagentVL2025.120.845——
Qwen2.5-VL2025.120.844——
ChameleonInput=Fix Res., RoPE=1D-RoPE, Backbone=Dense, Parameter Scale=8B2025.100.7——