Skip to content
Brief The AI-First Web ·

Alibaba releases Qwen4-Max, says it cuts the cost of a task by 60%

The new flagship model is closed, but Alibaba says smaller open versions will follow.

In brief
  • Alibaba released Qwen4-Max on November 16 and says the average task costs 60% less than on its previous model. The figure comes from an internal test.
  • The model is closed, with smaller open versions promised but not dated.

Alibaba released its Qwen4-Max AI model on November 16, Tech in Asia reported. Developers reach it through an API, a way to call the model from their own software. It also runs in Alibaba apps such as Qianwen Office, Qoder and Qianwen. Alibaba said the average cost of finishing a task is 60% lower than with its previous model.

The South China Morning Post, which Alibaba owns, reported that Alibaba gives the model 2.4 trillion parameters, the same as Qwen3.8-Max. Alibaba says only 71 billion work on each token, 25% fewer than before. A token is a small piece of text. Fewer parameters per piece means less computing, which the Post ties to the lower cost per task. Tech in Asia said cached requests start at 0.5 yuan, about US$0.075, per million tokens. The Post gives input pricing of the same US$0.075. The 60% comes from Alibaba's own Work360 test, the Post said. No outside test has been reported.

Qwen4-Max is closed. Alibaba said smaller open models will follow, with no dates given. The 60% is a vendor's own per-task result, so the meaning of "task" carries the claim. Buyers can ask which tasks Work360 covers, which older model is the baseline, and what the non-cached price is.

A WebPulse Brief: a short report of an important event, written by the WebPulse Newsroom with AI assistance and checked against the reporting below. When there is more to explain, we follow up with a full story. How we use AI.

Reporting: Alibaba Cloud, South China Morning Post.