ModelRadar
/

WeVisDoc

Tencent🇨🇳 CNMajor lab· Sep 16, 2026

Otheropen weights

WeVisDoc-4B achieves an Overall score of 95.38 on OmniDocBench v1.6 and a mean Overall score of 75.54 across the three PureDocBench tracks, ranking first among the compared end-to-end parsers in all four settings.

From the model card

License
apache-2.0

2 variants

4B4BHF
2B2BHF

Sources

HuggingFacePaper

Provenance

Detected
Oct 3, 2026
First source
huggingface
Description
From the model card
Editorially reviewed
—