Alibaba opens Qwen4-Max to subscribers only, through Nov. 25

Anyone who wants to try Qwen4-Max, Alibaba's new flagship model, has until Nov. 25 and a single way in: a Token Plan subscription, in its Personal or Team edition. Alibaba Cloud's documentation says pay-as-you-go is not supported and that the model will be retired when the preview ends. It lists no per-token price.

The Qwen account announced the preview on X on Tuesday night, Mexico time. Besides the Token Plan, the account offers early access in QwenWork, Qoder, the Qwen app and QwenStudio. The preview runs from 10:00 on Nov. 11 to 24:00 on Nov. 25, Beijing time. In Mexico City that is 20:00 on Nov. 10 to 10:00 on Nov. 25.

What the spec sheet says

The model ID is `qwen4-max-preview`. Alibaba calls it its next-generation flagship multimodal model, built for coding, office work, frontend design and agents.

• Context: 1 million tokens, the units a model uses to count text, with a maximum output of 393,216 tokens.
• Inputs: text, images and video.
• Reasoning: on by default, with three effort levels: `low`, `medium` and `xhigh`, the default.
• Tools: function calling, web search and structured JSON output.
• Compatibility: the API formats of OpenAI (Chat Completions and Responses) and Anthropic (Messages).

One catch for developers: in Chat Completions, `max_tokens` covers reasoning and the answer together, so a cap meant only for the answer can cut it off.

What hasn't been tested

Alibaba published no benchmark results and no model card. Any claim about the model's performance comes from the company, and there is no independent comparison with rival models. Nor does it say whether the Personal plan is open to users in Mexico; the API is served from Singapore.

The other tiers attributed to the family, such as Plus and Flash, don't appear in the official documentation. The one firm date is the end of the preview: Nov. 25.