WeVisOne-VL 4B

Tencent WeChat Vision's 4B vision-language model built on Qwen3-VL for object detection, grounding, counting, OCR, segmentation and alpha matting. Released under Apache 2.0 with a required 9.6M-parameter pixel decoder and a 262,144-token text-backbone configuration. Targets single-turn image perception with structured outputs; the pixel decoder does not support video grids. Available for self-hosting; no verified hosted token price is recorded.