Qwen4-Max-Preview Opens on Alibaba Token Plan

Alibaba Cloud has opened Qwen4-Max-Preview, which its documentation calls “the next-generation flagship multimodal model in the Qwen family,” through Token Plan only. The Qwen4-Max-Preview model page, last updated on 11 November 2026, says the model “is designed for coding, office productivity, frontend design, and complex agentic tasks,” combining text generation, image and video understanding, and deep reasoning. A matching entry appears on the Newly Released Models list, dated 11 November 2026, under the Singapore Token Plan scope.

The access terms are as important as the model name. “Qwen4-Max-Preview is currently in preview and is available through Token Plan only. Pay-as-you-go access is not supported.” The preview “runs from 10:00 on November 11, 2026 to 24:00 on November 25, 2026 (UTC+8)” — 02:00 GMT on 11 November to 16:00 GMT on 25 November — and “the model will be retired after the preview ends.” Alibaba also says “no changes will be made during this period.” The model ID is qwen4-max-preview.

The official pages are silent on three things readers often expect next. They give no parameter count, no API price, and no open-weight release. Those figures are not in the Alibaba Cloud documentation, and they are not supplied here from other outlets.

A 1M-token preview with thinking on by default

The published limits are large, and they come with a shared budget. Context length is 1 million tokens. Maximum output is 393,216 tokens, “shared by thinking and the final answer.” Thinking is enabled by default. Developers can set reasoning_effort to low, medium, or xhigh, and xhigh is the default. Because thinking and the answer share the same output budget, a limit sized only for the final reply can cut the response short.

The model accepts text, images and video, and the documentation lists function calling, web search, built-in tools, and structured outputs in JSON Object or JSON Schema. It is “compatible with OpenAI Chat Completions, Responses, and Anthropic Messages APIs,” so existing client libraries can be pointed at the Token Plan endpoint without a new protocol. Access is through Token Plan Personal or Token Plan Team, with a dedicated API key and a Token Plan base URL in the Singapore region. Alibaba points developers to those guides rather than to the usual pay-as-you-go Model Studio endpoints.

The Newly Released Models list files the preview under Singapore and Token Plan, dated 11 November, and repeats that it is “currently in preview” and “available through Token Plan only.” That is the whole availability story: a regional token-plan window, not a generally available Model Studio SKU, and not a checkpoint that can be downloaded when the window closes.

This is a different kind of Qwen announcement from the ones TSN has recently covered. Alibaba is already selling Qwen Glasses in China, a hardware product in the same family. Qwen4-Max-Preview is a cloud API, on a paid token plan, with a retirement date. It is also unlike the recent run of open Qwen derivatives, including Nous’s Arete-27B, which ships as a reworked Qwen checkpoint. And it sits at an angle to the earlier argument that open-source models such as Qwen3.8 now ship in months, with weights and serving stacks attached: this flagship preview discloses neither weights nor a lasting public endpoint.

A flagship that will be switched off

The accurate way to report this launch is narrower than “Alibaba has released Qwen 4 Max.” What opened on 11 November is a time-limited preview of a multimodal flagship, available only on Token Plan, under the ID qwen4-max-preview, with a 1-million-token context and a 393,216-token shared thinking-and-answer cap. Thinking is on by default, at xhigh unless the caller chooses otherwise. The official pages are silent on size, price and weights. After 16:00 GMT on 25 November 2026, Alibaba says the model will be retired.

That retirement date is the fact developers have to plan around. A two-week preview can be useful for trying coding, office, frontend and agent workflows, and for testing the OpenAI- and Anthropic-compatible endpoints. It is not a generally available model, and it is not an open-weight release. The facts to watch are whether Alibaba replaces the preview with a lasting Qwen4-Max endpoint, and whether any later official page adds a parameter count, a price or a weight drop. Until then, Qwen4-Max-Preview is a Token Plan trial with an end date.