Alibaba released Qwen4-Max on November 16, a proprietary mixture-of-experts model with about 2.4 trillion total parameters and a 1-million-token context window. It accepts text, image and video input and is available through the QwenCloud API, Qwen Studio, QwenWork and Qoder.

The model activates roughly 71 billion parameters per token, 25% fewer than its predecessor Qwen3.8-Max at the same 2.4T scale. Alibaba says this cuts the average cost per task by about 58% on its Work360 benchmark, while other coverage cites 62%, and it also lowers API prices.

The 2.4T figure and the savings are Alibaba's own claims and not independently verified. Unlike the open-weights Qwen3.8-2.4T-A95B, Qwen4-Max cannot be downloaded.