Qwen3.8 Max Weights Released with 2.4T Parameters and Custom License

Alibaba releases Qwen3.8-2.4T-A95B weights on Hugging Face and ModelScope, shifting to a custom license that restricts large-scale commercial use without approval.

Woofun AI reports that Alibaba has made the Qwen3.8-2.4T-A95B model weights available for download on Hugging Face and ModelScope. This release marks the first public availability of Qwen's Max-tier flagship weights, featuring 2.4 trillion total parameters with 95 billion activated per inference. The open-source version natively supports a 262K context length, extendable to approximately 1 million tokens, and requires 'Thinking' mode activation while accepting only text input. Unlike the cloud-based counterpart, it lacks visual input support, non-Thinking modes, default 1M context, and built-in tools. The licensing framework has shifted from Apache 2.0 to a custom Qwen license, permitting individual download and modification but mandating separate approval for large-scale commercial applications. The accompanying Qwen3.8-27B model remains unreleased.