Alibaba’s new AI accepts text, images and videos. Before trying it, check the deadline and access rules.
Qwen4-Max has arrived in preview with ambitious specifications and an important date: November 25, 2026. The period announced by Alibaba ends at 1 p.m. Brasília time that day. The qwen4-max-preview model will then be retired. This does not mean the end of the Qwen family, only this temporary version.
The documentation lists a context window of 1 million tokens, plus image and video understanding. Tokens are the small units into which AI divides content. The context window is the space available to work with that material. Source: Alibaba Cloud.
The Mais FPS take: A larger context window does not replace testing: evaluate the results before relying on a time-limited preview.
More space does not guarantee better answers
Imagine bringing together code, screenshots and documentation to ask for help with a website. A larger window may let you work with more references at once. But accepting a lot of information does not prove the model will find the right problem.
Mais FPS has not tested Qwen4-Max yet. This article contains no in-house comparison of its quality or speed against other AI models.
How much does it cost, and who can use it?
In Alibaba Cloud Model Studio, documented access is through the Token Plan, with no pay-as-you-go option during this preview. Access conditions.
Personal Lite normally costs US$6 per month, with 20,000 Credits. That is the service’s billing unit, not an equivalent number of tokens. Consumption varies; this is not unlimited access.
There is an important restriction: Personal allows interactive use in coding tools and agents, but prohibits automated scripts, servers for your own applications and batch calls without interaction.
The available infrastructure is in the Singapore region, with global processing. This does not confirm eligibility for every Brazilian account; subscription availability and the final price must be checked in the account. Personal pricing and rules.
Our take: try it on a familiar task, comparing results and the work needed to fix them. Just do not turn a time-limited preview into a permanent dependency. The number is impressive; usefulness has to show up in the work.
