Illustration, AI-generated from the story text. Tap a panel to enlarge.

Tencent opens a 5-billion-parameter model that parses documents and images with one encoder

Tencent posted Youtu-Parsing-Omni, a 5-billion-parameter model that reads documents, images and other formats into one schema, with a paper on Thursday and updated weights on Hugging Face on Friday. Tencent reports 96.96 on the OmniDocBench v1.6 document-parsing benchmark, which would be the top score, and the best open-model average on its own OmniParsingBench. The license is Tencent's own. It was the only new open-weight release from a large Chinese company we found this week, in a niche where Baidu shipped Unlimited-OCR-2 days earlier. SourcesAA

Sources

• AYoutu-Parsing-Omni paper (Hugging Face Papers)huggingface.co
• Atencent/Youtu-Parsing-Omni (Hugging Face)huggingface.co

This story ran in the Morning Brief of October 30th 2026.