Youtu-Parsing-Omni unifies parsing across four modalities
Youtu-Parsing-Omni is a 5B-parameter parser designed to convert documents, images, audio, and video into structured records grounded in space and time. It combines a shared-weight encoder, cross-modal fusion, a unified output schema, and self-evolving post-training across ten parsing tasks. The authors report a 96.96 overall score on OmniDocBench v1.6 and a 75.08 average on OmniParsingBench, while code and model weights are publicly available.
The release provides a publicly available 5B model and a unified architecture for parsing heterogeneous visual, textual, and audio inputs, which may be useful for agents that need to read, locate, and cite evidence across modalities.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2610.39301 (cs)
[Submitted on 29 Oct 2026]
Title:Youtu-Parsing-Omni: One Encoder, One Schema, All Modalities
Authors:Yunhang Shen, Zhangquan Chen, Xiaotian Li, Zhongpeng Cai, Qing Zhou, Zekai Zhang, Kun Zhang, Zezhong Wu, Jian Hu, Haihan Gao, Heting Gao, Xin Li, Yinsong Liu, Deqian Jiang, Di Yin, Haoyu Cao, Xing Sun, Bing Liu, Yanqiu Qu, Yunfei Wu, Kun Yin, Zhixiang Wei, Yongxiang Hua, Qianyu Li, Xiaoxing Hu, Peiyuan Zhang, Lijiang Li, Dong Xu, Zhenchuan Yang, Yiliang Jiang
View a PDF of the paper titled Youtu-Parsing-Omni: One Encoder, One Schema, All Modalities, by Yunhang Shen and 29 other authors
#Youtu-Parsing-Omni#Youtu-Omni-Encoder#OmniSchema#OmniSchema-Aware On-Policy Distillation#Tencent#OmniDocBench v1.6#OmniParsingBench