
Alibaba's Qwen 4 Max Leads China's AI Agent Race—But You Can't Download It
Alibaba has launched Qwen 4 Max, a 2.4-trillion-parameter AI model that leads Chinese competitors in agent rankings while maintaining closed weights.
Alibaba has released Qwen 4 Max, its most powerful AI model yet and the first flagship built on its new Qwen 4 architecture. The model features 2.4 trillion parameters and can process up to one million tokens of context, making it capable of handling lengthy codebases or extensive document collections.
Qwen 4 Max leads Chinese competitors on Arena's agent leaderboard, surpassing Tencent, Moonshot AI and DeepSeek in that particular test. The model uses a mixture-of-experts design, activating 71 billion parameters at a time compared to 95 billion in its predecessor, allowing for more efficient computing power usage.
Visual capabilities
The model's visual capabilities may be particularly notable for regular users. On Agent Arena, Qwen 4 Max placed 14th among individual models and made Alibaba the fifth-ranked laboratory overall. In Code Arena's web-development ranking, it debuted sixth globally at 1,750 points, showing significant gains in visual design.
Alibaba's demonstrations show the model building game environments and translating factory layouts into 3D models, though these are company-selected examples rather than independent tests. The visual capability focuses on understanding images and video to produce code or other text-based outputs rather than functioning as an image generator.
Cost and accessibility
Qwen 4 Max is priced at $1.50 per million input tokens and $4.50 per million generated tokens through QwenCloud, representing a 25% reduction from the previous flagship's rates. Reused input from an automatic cache costs just $0.075 per million tokens, potentially making repetitive workflows substantially cheaper.
Alibaba says its internal testing across hundreds of professional specialties found that the model used 35% fewer tokens and reduced average task costs by roughly 60% compared with Qwen 3.8 Max.
The significant limitation for the developer community is that Qwen 4 Max maintains closed weights. While developers can access it through an API or Alibaba's apps, they cannot download its internal parameters, modify it, or run it privately on their own hardware.
Market context
The launch occurs amid a competitive AI race in China, where Moonshot's Kimi K3 has pushed open-weight models toward Western performance, and DeepSeek has competed aggressively on API prices. While Qwen 4 Max leads some Chinese rankings, American labs still dominate the top spots in Arena's overall agent and web-development tables, though at considerably higher costs.
Alibaba has confirmed that smaller Qwen 4 models with open weights will be released in the future, allowing developers to build on this generation rather than just renting access to its largest model. The company has not yet announced the sizes or release dates for these smaller versions.