A model Alibaba only let some people try last week is now on general sale. Qwen4-Max, the company’s top Qwen model, has left the preview that opened on 11 November and can be bought through QwenCloud, plus the Qwen Studio, QwenWork and Qoder tools.
Size and limits first. The network has 2.4 trillion parameters in a mixture-of-experts layout, takes pictures as direct input, and allows requests of up to one million tokens, enough for a long report or a sizeable codebase in one go.
Then the money. Alibaba says Qwen4-Max burns fewer tokens than its predecessor Qwen3.8-Max to do similar work, and the API prices went down too. Put together, a typical task on Alibaba’s Work360 benchmark now costs RMB 11.16 where it used to cost RMB 26.58, a fall of roughly 58%. The benchmark is Alibaba’s own, so that saving stays a claim until others measure it.
That comparison is against Alibaba’s own earlier flagship, not against models from OpenAI, Google or Anthropic, so it says little about where Qwen4-Max ranks. Developers weighing a switch should check price, token count and image input, because those decide what a long, image-heavy job costs.
Alibaba did not say how much cheaper the API is per token, only the combined effect on its benchmark, which is why the 58% should not be read as a price cut of that size.
Two extras came with the launch. A plugin lets agents hand in a first draft and then revise it as the user reacts, and smaller Qwen4 models are slated for open-source release, with no date. The earlier preview was limited to a small group, and exact per-token rates sit on QwenCloud’s pricing pages.



