The race to build the most capable AI model just picked up another serious contender. On Sunday, November 15, Alibaba launched Qwen4-Max, the most powerful model in its Qwen family so far.
The company says the new model handles complex professional and scientific work better than its predecessor while using fewer resources and adding stronger safety protections. Here is what you need to know.
The headline specs
According to Alibaba’s announcement, Qwen4-Max has:
- 2.4 trillion total parameters, built as a “Mixture-of-Experts” model
- Only about 71 billion parameters active for each token, which keeps running costs down
- A one-million-token context window, enough to take in very long documents or large codebases at once
- Native vision, so it can work with images as well as text
The Mixture-of-Experts approach means the model doesn’t use all of its “brain” for every word. Instead, it routes each piece of work to the most relevant parts, which is a big reason Alibaba says it is cheaper to run.
Faster, cheaper, better
Alibaba says Qwen4-Max beats Qwen3.8-Max, released in August, while being more efficient. On its own Work360 benchmark, which covers challenging tasks across 721 specialized fields, the company reports that the new model uses 35% fewer tokens in its high-effort reasoning mode and cuts the average cost per task by around 60%.
The company also says Qwen4-Max ranked sixth on Code Arena, a real-world coding leaderboard, and scored well on tests for AI agents, workplace productivity and visual reasoning. As always with company-reported benchmarks, independent testing will tell the fuller story.
What it’s built to do
Alibaba pitches Qwen4-Max as an “AI teammate” for demanding work. “It can take on a wide range of creative and professional work, spanning 3D modeling, game development, precision manufacturing, finance, law, healthcare, and frontier scientific discovery,” the company said, as quoted by Reuters.
A few of Alibaba’s showcase examples:
- In a manufacturing demo, the model cut manual programming time for a machined part from about 475 minutes to 24 minutes.
- In a science test, it ran a continuous 48-hour agent workflow to help discover an experimentally validated coolant molecule for chip cooling.
- In an Alzheimer’s research experiment, it made 454 tool calls while designing and optimizing 24 candidate molecules.
A bigger focus on safety
Safety was a major theme of the launch. Alibaba said training included security checks alongside coding tests, helping reduce software vulnerabilities such as code injection, memory-safety flaws and broken access controls.
The company also said it strengthened safeguards against misuse in areas like cybercrime and hazardous materials, and that internal testing found no evidence of significant high-risk capabilities. Reuters noted that Chinese AI developers have been publishing more safety information with new models amid growing global concern about AI risks.
How to try it
Qwen4-Max is available through Alibaba’s QwenCloud via its API and token plans, and on Alibaba platforms including QwenWork and QwenStudio. The model’s weights have not been released. Alibaba says smaller open-weight models from the Qwen4 series will be released to developers soon.
Why it matters
Chinese tech companies are major players in AI, especially in open-weight models that developers can download and run themselves. Qwen4-Max shows the competition is now as much about efficiency and cost as raw power. If Alibaba’s numbers hold up, businesses may get top-tier capability for a fraction of the price, which puts pressure on every other AI lab.
For everyday users, the takeaway is simple: AI models are getting more capable and cheaper to run at the same time, and that trend is not slowing down.
Sources: Alibaba Cloud, Reuters via StreetInsider.
Image: Free computer server room image via Rawpixel, CC0 1.0