Alibaba launches closed-weight Qwen4-Max AI model
Alibaba launched Qwen4-Max on November 16, 2026, making its flagship artificial intelligence model available through hosted services while keeping its underlying weights closed. Developers can use it through the company’s QwenCloud API and apps, but cannot download the model to run or modify independently.
The model has 2.4 trillion parameters and activates 71 billion for each token processed, according to Alibaba. It accepts text, images and video and has a context window of up to one million tokens. Alibaba describes the model as built for coding and professional tasks, among other uses.
Independent leaderboard results show strong but not dominant placements. Qwen4-Max ranked 14th overall on Arena’s Agent Arena and was the highest-ranked Chinese model there. The score, 7.46 across 14,727 sessions, was preliminary. On Code Arena’s WebDev ranking, it placed sixth globally with a preliminary 1,750 points. The rankings measure performance in specific evaluations, not a general result across all AI tasks.
Alibaba’s own tests report improvements over its previous flagship, Qwen3.8-Max. On the company’s Work360 benchmark, which it says covers tasks across 721 specialized fields,...