Qwen4-Max preview: how to use it and what it costs
2026-11-11 · 5 min read
Qwen4-Max preview is Alibaba's first public look at its next flagship model, and you use it by buying a Token Plan subscription (from $6 a month, $3 for new accounts) and setting the model ID to qwen4-max-preview in a coding tool such as Claude Code, Cursor or Qwen Code. It runs for two weeks only: Alibaba's documentation says the preview opened on November 11 and the model will be retired at the end of November 25, China time.
The announcement came from @Alibaba_Qwen on X. I could not load the post itself, so I am not quoting it. The Neuron's digest, which links to it, says Alibaba teased the whole Qwen4 family and put the weight on coding and AI agents. TechNode's report the same day adds that the model is also in early use inside the Qwen app, Qoder, QwenWork and QwenStudio.
What Qwen4-Max preview can do
These numbers come from Alibaba's own model page and Token Plan docs, not from third-party testing.
- Context window: 1 million tokens
- Maximum output: 393,216 tokens, shared between the model's thinking and its final answer
- Inputs: text, images and video. Output is text.
- Thinking is on by default, with the effort setting at its highest level (
xhigh) unless you lower it - Works with the OpenAI Chat Completions, OpenAI Responses and Anthropic Messages API formats
Alibaba has not published benchmarks, a per-token price or open weights for this model, and it has not said what replaces it after November 25. The docs promise only that the model will not change during the window.
What it costs
There is no pay-as-you-go option. The preview is Token Plan only, and the Token Plan bills in credits instead of tokens. These are the Personal tiers listed in Alibaba's docs:
| Plan | Monthly price | Credits per month |
|---|---|---|
| Lite | $6 ($3 for new users) | 20,000 |
| Essential | $10 ($5 for new users) | 45,000 |
| Standard | $18 | 80,000 |
| Pro | $43 | 200,000 |
The new-user prices run until November 30 and apply only to accounts that have never bought a Personal plan. During the preview, Alibaba says Qwen4-Max calls burn credits at as low as 20% of the standard rate, and calls made between 22:00 and 08:00 China time drop to 8%. In US Central time that is 8 a.m. to 6 p.m., so the cheap window happens to cover the whole workday here.
Read the fine print before you pay. Personal plans cannot be cancelled, and unused credits do not roll over. Alibaba also does not publish how many credits a given request costs, so you will not know your real cost per task until you have run a few.
How to try it
- Sign up for a Token Plan Personal subscription on Alibaba Cloud Model Studio. Lite is enough for a test.
- Create a Token Plan API key. It is separate from a standard Model Studio key, and a pay-as-you-go key will not work.
- In your coding tool, point it at the Token Plan base URL shown in your console. Alibaba's pages list different URLs in different places, so copy the one your account shows.
- Set the model to
qwen4-max-preview. Your bill will show it asqwen4.0-max-preview, which is the same model. - Give it a real task: a bug in a large repo, or a long document plus a question that needs the whole thing in view.
The docs name Claude Code, Cursor, Qwen Code, Qoder and OpenClaw as supported tools.
The limits you should read first
The Token Plan terms allow interactive use inside coding and agent tools. Automation scripts, application backends and batch jobs are not allowed, and Alibaba says violations can get a key suspended. So you can sit in an editor and work with this model, but you cannot wire it into a product or a nightly workflow.
Personal plans are served from Singapore, and the docs say prompts and outputs cross borders for inference. None of the pages I read state whether your data is used for training. If you handle client records, keep them out of the test. Safety is also a fair question to ask of any new Qwen release: when Cohere looked at an earlier model, the findings were mixed, which I covered in Is Qwen3.8-Flash-Next safe?.
My read
This is a two-week trial, priced low so developers will bother. Alibaba has moved fast on coding before, including the time Qwen took a coding crown without a new version number, so I think the model deserves a look. But with no benchmarks, no price and no ship date, nobody can plan around it yet. If you want a new frontier model you can put into production this week, Gemini 4 Carbon went generally available on November 10 with published pricing.
For a business owner there is not much to do here. New models keep arriving as short previews with strings attached, so read the plan terms before you read the launch post. Three dollars is a cheap way to compare this one against the tool you already pay for, as long as the work you feed it is not sensitive.
If you are trying to work out which of these tools would save your team real hours, New Face Design offers a free process audit. We look at how the work moves today and tell you where automation would pay off and where it would not.