Alibaba's Qwen team just did something it has never done before: open-sourced its Max-tier flagship weights. Qwen3.8 is now available on Hugging Face and ModelScope as Qwen3.8-2.4T-A95B — a 2.4-trillion-parameter model that activates 95 billion parameters per inference.
This is also the first time Qwen has opened up a Max-level flagship. But don't expect an exact replica of the cloud-hosted Qwen3.8-Max. The official model card notes that the open version natively handles a 262K context window, expandable to around 1M, and lets you adjust reasoning effort. Right now, it's text-only, and Thinking mode is required. Cloud...