Qwen/Qwen3-VL-30B-A3B-Thinking

Qwen3-VL is the most powerful visual language model in the Qwen series to date. The model has undergone a comprehensive upgrade, featuring enhanced text understanding and generation, deeper visual perception and reasoning capabilities, extended context length, improved spatial and video dynamic understanding, and stronger agent interaction abilities. This "Thinking" version, with its enhanced reasoning capabilities, is built on a Mixture-of-Experts (MoE) architecture and excels at tasks such as interacting with graphical user interfaces of PC/mobile devices, generating code from images, and performing advanced multimodal reasoning in STEM fields. It natively supports a context length of 256K and boasts expanded OCR capabilities that cover 32 languages.