Tencent releases WeVisOne-VL for unified visual perception tasks

Tencent’s WeChat Vision team has released WeVisOne-VL, an Apache 2.0 vision-language model available in 2B and 4B versions. The models combine detection, referring, counting, pointing, OCR, segmentation and alpha matting. A 9.6M-parameter pixel decoder produces masks and mattes without requiring a separate SAM model. When a requested object is not present, the system can return None instead of generating a box.

Built on Qwen3-VL, WeVisOne-VL is aimed at fine-grained localization rather than open-ended conversation. It represents boxes and points with normalized coordinate tokens, while pixel tasks use an object token to query the decoder. Training involved more than 100M images and about 3.6B boxes, followed by three supervised fine-tuning stages and a final GRPO stage.

Tencent reports that the 4B model achieved 71.3 F1 on Dense200, 81.8 F1 on RefCOCOg test, 52.0 F1 on TotalText and 71.2 gIoU on ReasonSeg. These results are self-reported. The models are available through Hugging Face and can run with Transformers or vLLM in BF16; fine-tuning code is not yet available.