Alibaba Qwen4-Max launch cuts API prices by a quarter
The Alibaba Qwen4-Max launch opens the flagship model to pay-as-you-go API users with list prices 25% below its predecessor in Singapore, sharpening the price fight with U.S. labs.

Chinese AI labs are competing on price as well as performance. The Alibaba Qwen4-Max launch, which went live on Sunday evening U.S. time and on November 16 in Asia, makes the company's flagship model available through a metered API after a subscription-only preview, The Defiant and Unite.AI reported.
What happened
In Singapore, Qwen4-Max costs $1.50 for each million input tokens and $4.50 for each million output tokens, down from $2 and $6 for Qwen3.8-Max, a 25% cut on both, according to Unite.AI. The Defiant said the rate for implicit cache-hit input tokens is $0.075 per million, 70% below the older model. Prices differ in other regions, and Unite.AI noted a free quota of one million tokens in Singapore for 90 days.
The model first appeared on November 11 as a preview available only through Token Plan, Alibaba's subscription service for interactive tools, The Defiant said. The formal release adds pay-as-you-go access through QwenCloud and Model Studio, with availability listed in Singapore, Beijing, Frankfurt, Hong Kong, Tokyo and Virginia.
For developers, the change is mainly about access. Under the preview the model could only be reached through the subscription plan, which Unite.AI said bars automation scripts and backend use, so production apps had no clean way to call it. Metered pricing removes that hurdle, and explicit caching with five-minute or one-hour lifetimes can trim the bill for repeated prompts.
Why it matters
On specifications, Qwen's launch blog describes a mixture-of-experts design with a 2.4 trillion-parameter backbone that activates 71 billion parameters per token, The Defiant reported. It accepts text, image and video input, returns text, and handles a context window of one million tokens. Developers can call it through OpenAI- and Anthropic-compatible interfaces, with tool calling and structured outputs, which lowers the cost of switching from U.S. providers.
Alibaba also claims efficiency gains on top of the price cut. In its own tests, Qwen reported using roughly 48% to 49% fewer tokens on Terminal-Bench 4.0, according to The Defiant, though those figures have not been checked independently. Unite.AI said reasoning is on by default and shares a single output budget with the final answer, so long deliberation eats into response length.
The Alibaba Qwen4-Max launch comes as Chinese developers push hard on cost, following DeepSeek's open-source agent prototype. U.S. rivals are scaling revenue fast, with OpenAI reporting $10 billion in quarterly revenue, and cheaper Chinese models put pressure on their pricing for enterprise customers.
What's next
Smaller open-weight Qwen4 models are still to come, The Defiant said, which would widen access further. Independent benchmark results, rather than Alibaba's own evaluations, will decide whether the Alibaba Qwen4-Max launch wins developers away from American labs. The Defiant's report is here.
This article is for information only and is not investment advice.