XNews

Tencent WeChat Open-Sources WeVisDoc Document Parser

Tencent WeChat Open-Sources WeVisDoc Document Parser

Tencent WeChat open-sourced WeVisDoc, an end-to-end document parser converting page images to Markdown. It uses Qwen3-VL fine-tuned models and a two-stage training: coverage expansion then residual-driven capability allocation. On OmniDocBench it scored 95.38, and on degraded documents stage II added up to +4.03.

Source posts