List your company
ResearchFormerly WizWand

Multi-modal Document Understanding on OCRBench

270Recognition Score16bit
163.92191.46219246.54May 19, 2026
Updated 5mo ago

Evaluation results

MethodLinks
16bitModel=Qwen3-VL-8B, Precision=16-bit full-precision2026.0527018117418152858
OTTModel=Qwen3-VL-8B, Quantization=INT2, Group Size=1282026.0527017917017754850
OScaRModel=Qwen3-VL-8B, Quantization=INT2, Group Size=1282026.0527018017018353856
TurboQuant+Model=Qwen3-VL-8B, Quantization=2.5-bit2026.0526917816918249847
KIVIModel=Qwen3-VL-8B, Quantization=INT2, Group Size=1282026.0526817817318052851
RotateKVModel=Qwen3-VL-8B, Quantization=INT2, Group Size=1282026.0526716312716235754
QuaRotModel=Qwen3-VL-8B, Quantization=INT2, Group Size=1282026.0526617110415031722
16bitModel=Qwen3-VL-4B, Precision=16-bit full-precision2026.0526118317118651852
TurboQuant+Model=Qwen3-VL-4B, Quantization=2.5-bit2026.0526018317017540828
OScaRModel=Qwen3-VL-4B, Quantization=INT2, Group Size=1282026.0525918116918148838
OTTModel=Qwen3-VL-4B, Quantization=INT2, Group Size=1282026.0525618016717751831
KIVIModel=Qwen3-VL-4B, Quantization=INT2, Group Size=1282026.0525517716017546813
QuaRotModel=Qwen3-VL-4B, Quantization=INT2, Group Size=1282026.0525417314915938773
RotateKVModel=Qwen3-VL-4B, Quantization=INT2, Group Size=1282026.052541438413423638
16bitModel=LLaVA-v1.6-vicuna-7B, Precision=16-bit full-precision2026.05187160821070536
OTTModel=LLaVA-v1.6-vicuna-7B, Quantization=INT2, Group Size=1282026.0518715772970513
OScaRModel=LLaVA-v1.6-vicuna-7B, Quantization=INT2, Group Size=1282026.05178158821010519
QuaRotModel=LLaVA-v1.6-vicuna-7B, Quantization=INT2, Group Size=1282026.0517715261910481
KIVIModel=LLaVA-v1.6-vicuna-7B, Quantization=INT2, Group Size=1282026.0517315769890488
TurboQuant+Model=LLaVA-v1.6-vicuna-7B, Quantization=2.5-bit2026.0517315673990501
RotateKVModel=LLaVA-v1.6-vicuna-7B, Quantization=INT2, Group Size=1282026.0516815264890473