Alibaba opens Qwen4-Max as a 15-day preview with a 1M-token context window

Alibaba's Qwen team opened Qwen4-Max to developers on November 11 as a preview reachable only through the company's Token Plan, with the model due to be retired when the preview closes on November 25, according to the official model page.

The Qwen account framed the generation around four claims: deeper knowledge, richer world understanding, stronger coding and stronger agentic capability. What Alibaba did not publish is nearly as notable — no parameter count, no weight release, no per-token price and no benchmark results.

Qwen4-Max is a multimodal flagship that takes text, images and video as input and covers text generation, image and video understanding plus deep reasoning. Alibaba positions it for coding, office productivity, frontend design and complex agentic tasks. The headline number is a 1 million token context window, with a maximum output of 393,216 tokens shared between the model's thinking and its final answer.

Thinking is on by default at a default reasoning effort of xhigh, and developers can dial it down through the reasoning_effort parameter, which accepts low, medium or xhigh; the documentation warns against setting it alongside a separate thinking budget. Tool calling covers function calling, web search and built-in tools, while structured outputs support both JSON Object and JSON Schema.

The API surface is deliberately familiar: it is compatible with OpenAI's Chat Completions and Responses APIs as well as Anthropic's Messages API, so existing tooling should be able to point at the new model without a rewrite.

Access is closed and time-boxed. The documentation states plainly that pay-as-you-go is not supported, so an API key on its own is not enough, and the same preview also runs inside QwenWork, Qoder, the Qwen app and QwenStudio.

Pricing is credit-based rather than per token. During the promotion, credits are consumed at as low as 20 percent of the standard rate, and calls between 22:00 and 08:00 UTC+8 take a further 60 percent off, landing at 8 percent of standard. The model ID qwen4-max-preview carries the alias qwen4.0-max-preview, and bills show the alias.

The launch followed a cryptic tease. On November 6 the Qwen team posted a poster of birds gathering over a dark sea and forming the shape of a fish, with a classical Chinese line that stops one character short of the word Kun, plus three English words: "The becoming begins." No model name, date or specification appeared on it.

The fifteen-day window is the part worth planning around. A model that is switched off after two weeks is fine for evaluation and is unsuitable for anything shipping in production without a fallback already wired up. A wider Qwen4 release has no announced date and the remaining tiers of the family have not been detailed, so the next two weeks read less like infrastructure to build on than a limited audition of capability.