Qwen4-Max Preview: Alibaba’s 1M-Token Flagship, Until Nov 25
In short
- Alibaba opened a preview of Qwen4-Max, its new flagship model, with a 1M-token context and image and video understanding.
- It works only through a Token Plan, until November 25. No pay-as-you-go, and the docs list no price.
- It speaks the OpenAI and Anthropic API formats, so your existing tools can switch with one setting.
Alibaba just put its biggest Qwen on the table. The preview of Qwen4-Max started on November 11. It reads up to a million tokens at once, understands images and video, and plugs into the tools you already use. The catch? It disappears on November 25.
What happened
Qwen4-Max-Preview (model ID qwen4-max-preview) is available from 10:00 on November 11 until the end of November 25, Beijing time. After that it will be retired.
The specs from the official docs: a 1M-token context window and up to 393,216 output tokens, shared between thinking and the final answer. Thinking is on by default, with three levels: low, medium and xhigh. It understands screenshots, documents and video, calls tools, searches the web and returns structured JSON.
How do you get it? Only through a Token Plan, individual or team. Pay-as-you-go is not supported, and the docs page doesn’t give a price. Check the plan price in your QwenCloud account before you decide.
How to use it
- Swap it in. It supports OpenAI Chat Completions, OpenAI Responses and Anthropic Messages. Change the base URL and the model ID in your existing app or automation.
- Feed it whole documents. Contracts, supplier catalogs, long email threads, all in one request.
- Read screenshots and video. Dashboards, invoices, product demos.
- Race it. Run the same 20 real tasks through your current model and this one. Keep what wins.
Do the math: say you review 20 long documents a month, 2 hours each, at €15 an hour. That’s €600 a month. If a model reads and flags them in minutes, even a plan at €100 pays for itself. Those numbers are my assumptions, because the plan price isn’t published on the page.
Two tips from the docs. Allow extra room in max_tokens, because thinking tokens count against it. And don’t set reasoning_effort and thinking_budget together.
What it means
Chinese labs keep copying the US playbook and the US API formats. Compatibility with OpenAI and Anthropic means switching costs are close to zero. That puts pressure on prices.
But it’s a preview with a hard end date. Don’t build a product on it. Test it, measure it, move on. I found no independent benchmarks, so treat every quality claim as unproven.
Read also: Apple Picks Alibaba’s Qwen for Apple Intelligence in China · Grok API Now Watermarks Every Text Output by Default · Kimi K3.1 Pro Weights Are Out: 2.8T Parameters, Who Should Care
Ivan’s Take
Worth a two-week test, not a migration. A 1M context and zero switching effort make it cheap to try. I’d check the plan price first and only pay if you have real long-document work waiting.
Source: QwenCloud
HawkValue
If it doesn’t save you time or money, you won’t see it here
- Facts, not hypeEvery price and number is checked against the official source
- Always the real priceYou see what it costs before you click
- I say when it’s weakEven when it’s OpenAI or Google
