Screenshots show the official pages on the capture date and may differ from the pages at publication. Select an image to enlarge it.
On November 11, 2026, Alibaba released "Qwen4-Max-Preview," a trial version of the top model in its next-generation "Qwen4" AI lineup. According to Alibaba Cloud, it is the next flagship model in the Qwen series, built for programming, office work, web page design and complex AI agent tasks. Alibaba revealed in September that Qwen4 was in training, but this is the first time it has been released in a form people can actually use.
Qwen4-Max-Preview specs
According to Alibaba Cloud's developer documentation, the main specs are as follows.
- The amount of text it can handle at once (its context) is 1 million tokens
- A single output can be up to 393,216 tokens, including both its thinking process and the final answer
- In addition to text, it can take in images and video
- Thinking before answering is turned on by default, and you can choose from three levels of thinking depth
- It supports calling external tools and web search
- It can be called in the same format as the OpenAI and Anthropic APIs
On the other hand, benchmark scores comparing its performance and the model's size (number of parameters) have not been published.
Available only until November 25
The availability period runs from 10:00 on November 11 to 24:00 on November 25, China time (from 11:00 on November 11 to 1:00 on November 26, Japan time). Alibaba Cloud says it will not change the model during this period and will stop offering it once the period ends. It has not said when a final version will follow.
To use it through the API, you need to sign up for Alibaba Cloud's monthly "Token Plan." It is not available under normal pay-as-you-go billing. The individual Token Plan comes in four tiers, from Lite at $6 a month to Pro at $43 a month, and during the period Qwen4-Max-Preview usage is reduced to at most 20% of the normal rate, and to 8% from 22:00 to 8:00 the next morning China time (23:00 to 9:00 the next morning Japan time). However, use is limited to interactive use from programming and agent tools such as Claude Code, Cursor, Qwen Code and Qoder; automated scripts and use behind the scenes in your own apps are not allowed. It is offered only in the Singapore region.
According to media reports, besides the Token Plan, it is also available early in the Qwen app, QwenWork, the AI-powered development tool Qoder, and QwenStudio.
What has not been announced yet
The Qwen4 lineup Alibaba outlined in September included four names, Max, Plus, Flash and 27B, but only a trial version of Max has been released this time. Alibaba has not yet announced when the other models will come out, how much the final version will cost, or whether it will release the models as open weights that anyone can download and use.
CoAI's take
Until now, Alibaba has released many Qwen models as open weights, and they have been used by developers around the world. This time, access is limited to two weeks and narrowed to interactive use from programming tools, which suggests Alibaba is watching how the model is actually used to polish it before an official launch.
For people trying it from Japan, the Token Plan is relatively cheap, and once you set an API key and endpoint, the model can be called from Claude Code or Cursor, which makes it easy to get started. However, Alibaba Cloud says that what you enter and the model's responses may be transferred across borders. Anyone handling internal code or confidential information should check their organization's policies before deciding to use it.
Sources
- Qwen4-Max-Preview (Alibaba Cloud Model Studio)
- Token Plan (Personal) overview (Alibaba Cloud Model Studio)
- Token Plan (Team) overview (Alibaba Cloud Model Studio)
- Alibaba's Qwen4-Max Preview Offers a 1M-Token Context Window, but Access Ends Nov. 25 (XenoSpectrum)
- Alibaba Cloud's 2026 Apsara Conference: Full-Stack AI Roadmap Along with Global Market Expansion Plan (Alibaba Group)
- Qwen4 family revealed for the first time: Max, Flash, and Plus launch simultaneously across three lines, with Qwen4 27B making an appearance (KoCPC)
