Alibaba Cloud brings Qwen4-Max to enterprise teams

Alibaba Cloud has announced Qwen4-Max for enterprise teams, positioning the model as an AI colleague for long, tool-heavy work that produces usable deliverables. The Qwen4 flagship has 2.4 trillion total parameters, native vision and a context window of up to 1 million tokens.
- Qwen4-Max has 2.4 trillion total parameters and a 1M-token context window.
- Live Delivery provides usable versions of long tasks as work progresses.
- Alibaba Cloud reports a 52.69 Qwen Work360 score and 58% lower average task cost.
- Launch demonstrations covered manufacturing, science, legal work and finance.
- Enterprise teams can contact Alibaba Cloud to plan a regional pilot.
Built for long-running professional work
Alibaba Cloud says Qwen4-Max is designed to work across hours, dozens of tools and thousands of pages, rather than stopping at a single answer. The model has 71 billion active parameters per token and an additional 328 billion N-gram embedding parameters.
Its architecture combines Gated DeltaNet layers with Qwen Sparse Attention to make long contexts more efficient. Qwen4-Max can also read images natively. A feature called Live Delivery provides a usable first version of a task early, then continues producing improved versions while people review and redirect the work.
Alibaba Cloud says the model is intended for manufacturing engineering, financial research, legal due diligence, scientific discovery with lab automation, 3D modeling, game development and cloud resource scheduling.
Launch demonstrations cover manufacturing, science and finance
In a manufacturing demonstration, Qwen4-Max built 3D production lines from 2D layouts, generated and simulated CAM toolpaths, and identified over-cuts and under-cuts before machining. In the case shown, hands-on programming time per part fell from about 475 minutes to 24, which Alibaba Cloud describes as about 95% less.
During a 48-hour autonomous science run with an integrated computational and robotic wet lab, the model built a 2.7-million-molecule candidate space, completed four design, synthesis and testing loops, and produced a data center coolant molecule that was validated in the lab as a real sample.
For IPO due diligence, Qwen4-Max sorted 198 unorganized files into 14 folders and delivered an 86-page report covering 12 modules and 14 legal issues, with conclusions cited to source files. In finance, it used 90 MCP tools to gather market data and produced an Excel workpaper and editable PPT report with figures linked to their sources.
Benchmark gains and enterprise access
On Alibaba Cloud's Qwen Work360 evaluation, which covers 26 top-level domains, 182 sub-domains and 721 fine-grained domains, Qwen4-Max scored 52.69, compared with 47.74 for Qwen3.8-Max. Alibaba Cloud says it used about 24% and 35% fewer tokens at Xhigh and High reasoning effort, respectively, and had a 58% lower average cost per Work360 task.
The company also reports that the model reduced model-call scheduling cost for data labeling by 10.42% on average against a reference baseline. On Last Secure Code Bench, the function-level rate of code that was both functional and secure rose from 22.0% for Qwen3.8-Max to 31.8%.
Enterprise teams can contact Alibaba Cloud to plan a Qwen4-Max pilot on their own workflows. Availability, features and pricing vary by region. Alibaba Cloud cautions that demonstration and benchmark results depend on the task, tools, data and configuration, so a scoped pilot is the way to assess performance for a specific workflow.


