Overview
WeVisOne-VL-2B is a 2-billion-parameter vision-language model from Tencent's WeChat Vision team, designed to unify object-level and pixel-level visual perception. Its released weights include a required pixel decoder of approximately 9.6 million parameters. (github.com)
Capabilities and uses
The model supports object detection, referring-expression localization, pointing, counting, OCR and text localization, segmentation, and image matting. Outputs can include points, bounding boxes, text, binary masks, and continuous alpha mattes. It is intended for visual perception research and related image-processing workflows. (github.com)
Limitations and licensing
The model targets single-turn image perception; its pixel decoder rejects video grids. Fine matte boundaries can be sensitive to image resolution, severe occlusion, complex transparency, and prompt wording. It is not intended to identify individuals or make autonomous high-impact decisions. Tencent releases the model under Apache 2.0, while third-party components retain their respective licenses. (huggingface.co)
Release date
October 23, 2026: the official model repository records the release of the weights and model card on this date. Its initial repository commit was on October 22. (huggingface.co)
Pricing
As checked on November 16, 2026, model-specific official input and output API prices could not be verified in Tencent Cloud's domestic CNY or international USD TokenHub pricing pages. No price or billing unit is assumed. Self-hosting costs depend on the infrastructure used. (intl.cloud.tencent.com)