Qwen4-Max-Preview

Qwen4-Max-Preview is Alibaba’s AI model preview for working with text, images and video. It returns text and can carry out reasoning and coding tasks. Alibaba opened the preview on 11 November 2026.

The preview has a fixed end date

Access is through Token Plan. The announced preview runs from 11 November at 10:00 to the end of 25 November 2026 in UTC+8, equivalent to 16:00 UTC on 25 November. Alibaba says this preview model will retire afterwards. Work that depends on it needs a migration plan before that deadline.

The Model Studio documentation separates this access from ordinary pay-as-you-go model billing. No per-million-token price for this preview is established in the published access terms. Check the current Token Plan allowance and supported account region before starting an evaluation.

Thinking and the answer share the output allowance

The documentation lists a one-million-token context window and an output budget of 393,216 tokens shared by thinking and the final answer. Tokens spent reasoning reduce the amount left for the answer. This is especially relevant to long coding jobs or responses containing many files.

Supported thinking levels are low, medium and xhigh, with xhigh as the default. Developers can compare the available settings for response time, output length and completion of their own tasks. A larger budget describes capacity; it does not guarantee a correct answer.

Compare it on the tasks you need

The preview accepts images and video as well as text. Useful checks include whether it follows file instructions, handles visual details accurately and completes tool-driven work within the available budget. The hosted preview terms are the basis for access; the opened documents do not establish downloadable Qwen4 weights or an official parameter count.

The cover is a conceptual illustration of model use, rather than a product screenshot.