Tencent
Nov 4, 2026

Tencent releases open-source WeVisOne-VL to handle detection, OCR, segmentation and matting in one model

Tencent’s WeChat Vision team has released WeVisOne-VL, an open-source vision-language model offered in 2B and 4B parameter versions. The model is designed to cover a wide range of visual tasks within a single system, including object detection, referring expression tasks, counting, pointing, OCR, segmentation, and alpha-channel image matting. According to the source, it does not require a separate Segment Anything Model, or SAM, to perform those functions. Its pixel-level output is produced by a 9.6 million-parameter decoder. MarkTechPost said the model was tested across more than 30 benchmarks and outperformed baseline models of similar size on most tasks. The release highlights Tencent’s latest work in multimodal vision systems and focuses on combining several image understanding and pixel-generation capabilities into one open-source model.

450
Tencent releases open-source WeVisOne-VL to handle detection, OCR, segmentation and matting in one model