Alibaba has launched Qwen4-Max, the new flagship of its Qwen family: a 2.4-trillion-parameter mixture-of-experts model with a one-million-token context window. The release immediately topped developer charts, leading Hacker News discussion on its launch day.

Alibaba prices access at $1.50 per million input tokens and claims the model cuts the cost of completing a typical task by about 60% compared with its predecessor. For companies buying AI by the token, that cost curve matters as much as benchmark scores.

The notable break with tradition is openness. Qwen built its reputation on releasing strong open weights that anyone could download and run. Qwen4-Max keeps its weights closed — a sign that even the most open major labs are starting to hold their largest models back.

The background

The Qwen family earned its place in the AI ecosystem the open way. Release after release, Alibaba published weights that developers could download, fine-tune and run on their own hardware, and the models became default building blocks for startups and researchers — especially outside the United States. That openness was a strategy: influence and adoption first, revenue later.

Qwen4-Max is built on a different scale of ambition. A 2.4-trillion-parameter mixture-of-experts design activates only a fraction of its parameters for any given token, which is how such a large model can answer at practical speeds. The million-token context window means entire codebases or book-length documents fit into a single conversation.

Why it matters

The pricing tells one story: at $1.50 per million input tokens with a claimed 60% cut in task cost, Alibaba is competing on economics as much as intelligence. For companies that buy AI by the token, a model that is good enough and far cheaper changes purchasing decisions more than a few benchmark points do.

The closed weights tell another. The industry's centre of gravity is shifting: as frontier models become more valuable — and more expensive to train — even the labs that built their reputations on openness are reserving their largest systems. Developers who built on open Qwen releases now have to decide whether the flagship's performance is worth depending on an API they do not control.

The wider story

Qwen4-Max also lands in the middle of a geopolitical contest. Chinese labs are releasing models that rival Western ones on capability while undercutting them on price, even as export controls restrict their access to the most advanced chips. A closed flagship from Alibaba does not end that contest — it changes its terms, from open weights anyone can copy to hosted services anyone can buy.

What happens next

Watch whether Alibaba keeps future mid-size Qwen releases open while reserving the flagships, and how developers respond: the open ecosystem may rally around the last open generation, or migrate to the API. The answer will say a lot about where the open-model era is heading.

Common questions

What is Qwen4-Max?

Qwen4-Max is Alibaba's new flagship AI model: a 2.4-trillion-parameter mixture-of-experts system with a one-million-token context window. It is priced at $1.50 per million input tokens, and Alibaba claims it cuts the cost of a typical task by about 60%.

Are Qwen4-Max's weights open?

No. Unlike earlier Qwen releases, which built the family's reputation on open weights anyone could download, Qwen4-Max keeps its weights closed. Access is through Alibaba's hosted service.