Alibaba Launches Qwen4-Max Preview, Restricted to Token Plan Until November 25
On November 11, Alibaba launched the preview of Qwen4-Max, the first model in its Qwen4 generation: a one-million-token context window, text, image, and video inputs, and API access restricted to the Token Plan subscription until November 25. No benchmark results have been released.
The Qwen4 generation kicks off with a trial run. This Wednesday, November 11, Alibaba rolled out the preview of Qwen4-Max, the flagship model of its upcoming Qwen family. The announcement was shared on X at 3:01 a.m. Paris time under the headline “Qwen4 is coming.” Developers can call it via API, but only with a QwenCloud Token Plan subscription, and only until November 25.
What is available
According to QwenCloud documentation, Qwen4-Max accepts text, image, and video inputs, features a one-million-token context window, and can generate up to 393,216 tokens in output, including reasoning. It supports function calling, built-in tools, and structured outputs. QwenCloud bills it as its “highest-quality” model, outperforming Qwen3.8-Max.
The API identifier is qwen4-max-preview (alias qwen4.0-max-preview). The Token Plan page sets the preview window from November 11 at 10:00 a.m. to November 25 at 12:00 a.m. Beijing time (UTC+8)—equivalent to November 11 at 3:00 a.m. to November 25 at 5:00 p.m. Paris time. The model will remain frozen throughout this window, after which it will be deprecated. It is not offered through standard pay-as-you-go pricing outside the subscription.
Beyond the API, Alibaba is providing early access inside its own product ecosystem: QwenWork, the coding tool Qoder, the Qwen app, and QwenStudio. The announcement does not explain how to select the model within each app.
A subscription gate
The Token Plan is QwenCloud’s subscription package, designed specifically for coding tools and agents. The pricing page advertises individual Lite and Essential plans starting at $6 per month, with an extra 50% discount during the ongoing promotion. During the promotional run, Qwen4-Max API calls consume just 20% of the standard rate in credits, dropping to 4% between 10:00 p.m. and 8:00 a.m. Beijing time (3:00 p.m. to 1:00 a.m. Paris time). QwenCloud has not disclosed Qwen4-Max’s standard credit cost or its eventual pay-as-you-go pricing.
What is missing
Alibaba touts “deeper knowledge,” a “richer understanding of the world,” and enhanced capabilities in coding and agents for front-end development and software workflows. Yet the announcement includes no benchmarks, no model card, no parameter count, and no licensing details. There is also no indication of whether other Qwen4 sizes are on the way or in what format.
Independent evaluations are similarly absent. On Arena’s text leaderboard, accessed on November 11, Alibaba’s top-performing model remains Qwen3.8-Max, ranked 23rd with a score of 1,484, trailing Gemini 4 Carbon (1,526). Qwen4-Max is not yet ranked.
Our take
Alibaba is opting for a brief, static, paid preview rather than a full-scale launch. For developers, that brings real value: the model won’t shift for two weeks, making tests reproducible across real-world workflows. Yet locking the API behind the Token Plan turns the subscription into a mandatory gatekeeper, and the scheduled November 25 shutdown serves as a reminder that this specific build will not remain in production. Without benchmarks or a technical report, it is impossible to gauge how Qwen4-Max stacks up against Gemini 4 Carbon, Claude, or GPT-6; claims regarding coding and agents remain unverified. The genuine test will come from public evaluations like Arena, as well as the final production release, for which neither pricing nor launch dates have been revealed.
Sources
- Qwen (X) — Qwen4 is coming, Qwen4-Max is now in preview on Alibaba's Token Plan Primary source
- QwenCloud — Token Plan individual (documentation, supported models) Primary source
- QwenCloud — Text generation and visual understanding (documentation) Primary source
- QwenCloud — Token Plan (pricing page) Primary source
- Arena — Text Arena, overall leaderboard Primary source — Accessed November 11, 2026.
- TechNode — Alibaba opens Qwen4-Max preview with a million-token context window Secondary source
A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.
