Qwen 3.8 for the first time. But... Load

Query formally released the Max-class model weight of Qwen 3.8, which is on-line Hugging Face and Modelscope, known as qwen 3.8-2.4t-a95b, with a parameter of 2.4t and activated by every reasoning 95b. This is the first time that qwen has opened the Max class flag weight. The open version supports the 262k context, which can be extended to approximately 1 m, and the intensity of the reasoning. The current version only supports text input and needs to start thinking mode. In addition, the qwen 3.8-27b weight has been opened, using a 27b dense structure, which supports photo and video understanding in the context of 262,144 token and can be extended to approximately 1 million token. After quantification, qwen 3.8-27b can operate locally on about 17 gb Ram or visible, suitable for personal developers。