According to CoinMeta, Qianwen has officially released the Max level model weights for Qwen3.8. Models Hugging Face and Modelscope have been launched, with the model name being qwen3.8-2.4t-a95b and a parameter size of 2.4T. Each inference activation consumes 95B of resources. This is the first time that qwen has opened up its Max level flagship weights. The open version natively supports up to 262K contexts and can be extended to about 1M, while also supporting the adjustment of inference intensity. The current version only supports text input and requires the activation of thinking mode. In addition, Qianwen has also made available the qwen3.8-27b weights, which feature a 27B dense architecture and natively support image and video understanding. The context capacity reaches 262,144 token, and can be extended to about 1 million token. After quantization, qwen3.8-27b can run locally on approximately 17GB of ram or video memory, making it suitable for individual developers to use.