Qwen4-Max: Alibaba’s new model costs 25% less, what changes

On November 16, 2026, Alibaba made Qwen4-Max available to everyone. It is the new top model in the Qwen family. Since November 11 it had been in preview, limited to people with a Token Plan subscription. Now you can also pay as you go through the API.

For people who use AI at work, the price is the main news: the list price is 25% lower than the previous model, Qwen3.8-Max. The rest of the announcement is made of figures claimed by Alibaba, and they should be read that way.

What is Qwen4-Max?

It is a model that reads text, images and video and answers with text. It uses a MoE design (Mixture of Experts): it contains a huge number of specialized parts, but only a small fraction of them works on each word. You can set how much it "thinks" before answering, on three levels. These are the figures stated by the company and listed in the documentation:

FeatureQwen4-Max
Parameters2.4 trillion in total, 71 billion active per word
InputText, images and video
Context1 million tokens
Maximum answer131,072 tokens (262,144 including the thinking)
Thinking levelslow, high (default) and xhigh
Downloadable weightsNo: cloud only
Where to use itQwenCloud (API or Token Plan), QwenWork, Qoder and Qwen Studio

How much does it cost compared to before?

The prices below come from the Alibaba Cloud Model Studio price list for the Singapore region, in US dollars per million tokens. A token is a piece of a word: one million tokens is several hundred pages.

  • Input text: $1.50 (Qwen3.8-Max: $2);
  • Output text: $4.50 (Qwen3.8-Max: $6);
  • Cache reads, meaning text you already sent and are reusing: $0.075;
  • In the other regions, Frankfurt among them: $1.413 for input and $4.24 for output;
  • Trial quota in Singapore: 1 million free tokens, valid for 90 days.

Alibaba adds that the model uses fewer tokens to do the same job. On its internal Work360 test, which simulates professional tasks from 26 fields, the average cost per task goes from $4.43 to $1.67 according to the company: 62% less. The score on the same test goes up from 47.74 to 52.69. Alibaba built this test itself, so there is no independent check yet.

Can you download it and run it locally?

No. Qwen4-Max runs only on Alibaba's servers, and the documentation also rules out custom training (fine-tuning) and installing it on your own servers. The announcement does promise that smaller Qwen4 models will be released as open source, with no dates or names. If you want an open model today, look at Inkling 2 Small, which we covered a few days ago.

One thing is open, though: the Qwen-Live-Delivery plugin. It is for agents, the programs that carry out long tasks on their own. With this plugin the agent hands over a first version of the work early and improves it in stages, so you can correct its course without waiting for the end.

What changes for people who work with AI?

If you only use ChatGPT, Claude or Gemini in the browser, almost nothing. Qwen4-Max matters to those who pay for models by usage: companies with automations, developers, anyone who has large amounts of documents processed. For them a 25% cut in the list price, plus the lower token use, shows up on the monthly bill. Moving from another provider is fairly simple, because the API accepts the same formats as OpenAI and Anthropic.

The point to weigh carefully is data. Everything you send is processed on Alibaba Cloud servers. Frankfurt is one of the available regions, but before sending contracts or client data, ask whoever handles privacy in your company. We would try it first on non-confidential material, comparing cost and quality with the model you already use. If you still have to choose that one, start from our guide ChatGPT, Claude or Gemini: which one to choose for work.

Frequently asked questions

Is Qwen4-Max free?

The API is pay as you go, with a trial quota of 1 million tokens in the Singapore region. The model is also available in Qwen Studio: the announcement does not give limits or terms of use for that version.

What happens to the preview version?

Qwen4-Max-Preview stays active only for Token Plan subscribers until November 25, 2026, then it is retired. If you use it, you have to move to the final model, which is called qwen4-max in the API.

What is the difference between Token Plan and the API?

Token Plan is a subscription with credits, meant for using the model inside coding tools. The pay-as-you-go API is for connecting it to your own programs, and you pay for the tokens you use.

Official sources

This article was written with the help of artificial intelligence models, starting from the official sources listed, and is reviewed before publication.

Condividi questo articolo sui tuoi socialShare this article on your social networksComparte este artículo en tus redes socialesTeile diesen Artikel in deinen sozialen Netzwerken

No products were found matching your selection.

Iscriviti alla newsletter di You vs AISubscribe to the You vs AI newsletterSuscríbete al boletín de You vs AIAbonniere den Newsletter von You vs AI