×
Community Blog Alibaba Cloud Brings Qwen4-Max to the Enterprise: An AI Colleague That Plans, Builds and Delivers Finished Work

Alibaba Cloud Brings Qwen4-Max to the Enterprise: An AI Colleague That Plans, Builds and Delivers Finished Work

Built on the new Qwen4 architecture with 2.4 trillion parameters, native vision and a 1M-token context window, Qwen4-Max takes on long, tool-heavy pro.

5eecdaf48460cde545f478b7a11f0b618b9b483b

Qwen4-Max for Enterprise.

Alibaba Cloud Brings Qwen4-Max to the Enterprise: An AI Colleague That Plans, Builds and Delivers Finished Work

Built on the new Qwen4 architecture with 2.4 trillion parameters, native vision and a 1M-token context window, Qwen4-Max takes on long, tool-heavy professional work at 58% lower average cost per task than Qwen3.8-Max on the Work360 evaluation.

SAN FRANCISCO, November 17, 2026: Alibaba Cloud today announced Qwen4-Max for enterprise teams. Qwen4-Max is the flagship model of the Qwen4 series: an always-on AI colleague that works, creates and explores alongside people. It is built to use professional tools, carry long tasks through to completion and deliver finished work, not just answers.

The work that matters inside an enterprise rarely fits in a single prompt. It runs for hours, across dozens of tools and thousands of pages, and it has to end in something a team can use. Qwen4-Max is built for that kind of work.

Qwen4-Max has 2.4 trillion total parameters, with 71 billion active per token and an additional 328 billion N-gram embedding parameters. Its new architecture pairs Gated DeltaNet layers with Qwen Sparse Attention to keep very long contexts efficient. It reads images natively and holds up to 1 million tokens of context. With Live Delivery, it ships a usable first version of a long task early, then keeps delivering improved versions, so teams can review and redirect the work while it runs.

In launch demonstrations, Qwen4-Max delivered finished work across industries.

Manufacturing. Qwen4-Max built 3D production lines from 2D layouts, generated and simulated CAM toolpaths, and caught over-cuts and under-cuts before machining. In the case shown, hands-on programming time per part fell from about 475 minutes to 24, about 95% less.

5eecdaf48460cde545f478b7a11f0b618b9b483b

In simulation, Qwen4-Max flags a local overcut caused by an oversized tool before any metal is cut. Official Qwen4-Max demo.

5eecdaf48460cde545f478b7a11f0b618b9b483b

Hands-on CAM programming time per part, in the case shown.

Science. In a 48-hour autonomous run with an integrated computational and robotic wet lab, Qwen4-Max built a 2.7-million-molecule candidate space, ran four design, synthesis and testing loops, and produced a new data center coolant molecule, validated in the lab as a real sample.

5eecdaf48460cde545f478b7a11f0b618b9b483b

Qwen4-Max schedules the robotic wet lab that synthesizes and tests its candidates. Official Qwen4-Max demo.

Legal. For IPO due diligence, Qwen4-Max sorted 198 unorganized files into 14 folders, cross-checked public records and delivered an 86-page report covering 12 modules and 14 legal issues, with every conclusion cited to its source file.

Finance. Using 90 MCP tools, Qwen4-Max gathered market data, cross-checked public sources and delivered an Excel workpaper and an editable PPT report in which every figure traces back to its source.

5eecdaf48460cde545f478b7a11f0b618b9b483b

198 files in, one cited report out.

Cloud operations. On Alibaba Cloud resource scheduling, Qwen4-Max reduced model-call scheduling cost for data labeling by 10.42% on average against the reference baseline, and reached the optimal schedule in offline resource rescheduling more than 14.61 times faster.

Qwen4-Max is also more efficient than its predecessor. On the Qwen Work360 industry benchmark, which spans 26 top-level domains, 182 sub-domains and 721 fine-grained domains, it scores 52.69, up from 47.74 for Qwen3.8-Max, while using about 24% and 35% fewer tokens at Xhigh and High reasoning effort. Average cost per Work360 task is 58% lower.

5eecdaf48460cde545f478b7a11f0b618b9b483b

Average cost per task on the Qwen Work360 evaluation, Qwen4-Max vs Qwen3.8-Max.

Security is designed in. Code-security checks were added to software engineering training. On Last Secure Code Bench, the function-level rate of code that is both functional and secure rose from 22.0% for Qwen3.8-Max to 31.8%, and the share of secure code generated at function, file and repository level at once rose from 8% to 14.7%. Getting started. Enterprise teams can talk to Alibaba Cloud today to plan a Qwen4-Max pilot on their own workflows. Availability, features and pricing vary by region. Get started: talk to our team.

Frequently asked questions

What is Qwen4-Max?

Qwen4-Max is the flagship model of the Qwen4 series. It has 2.4 trillion total parameters with 71 billion active per token, native vision and a 1M-token context window, and it is built for professional tool use, long-horizon tasks and delivery of finished work.

What does "Qwen4-Max for Enterprise" mean?

It is how Alibaba Cloud brings Qwen4-Max to enterprise teams. It is the same Qwen4-Max model, not a separate edition.

What is new compared with Qwen3.8-Max?

A new architecture (Gated DeltaNet with Qwen Sparse Attention, Gated Residual and N-gram embedding); a higher Qwen Work360 score (52.69 vs 47.74); fewer tokens per task (about 24% and 35% fewer on Work360 at Xhigh and High reasoning effort, about 48% and 49% fewer on Terminal-Bench 4.0); a 58% lower average cost per Work360 task; and more secure code (31.8% vs 22.0% function-level functional and secure pass rate on Last Secure Code Bench).

What kinds of work does it handle well?

Long, tool-heavy professional work: manufacturing engineering such as line layout, CAM toolpaths and simulation; financial research; legal due diligence; scientific discovery with lab automation; 3D modeling in Blender and game development in Unreal Engine; and cloud resource scheduling.

Are the launch results typical?

They come from specific launch demonstrations and are described as "in the case shown". Results depend on the task, the tools and the data. A scoped pilot on your own workflow is the best way to measure impact.

What is Live Delivery?

A way of running long agent tasks. Qwen4-Max delivers a usable first version early, then keeps delivering improved versions, so you can check progress and steer the work while it runs. The Qwen-Live-Delivery-Plugin is available alongside the model.

Can it work with images?

Yes. Qwen4-Max has native vision.

How is safety handled?

Qwen model governance follows six AI principles across the model lifecycle: people-centered, inclusive and honest, privacy-protecting, safe and reliable, trustworthy and controllable, and open governance. Code-security checks were added to training. For cybersecurity, data for defensive tasks such as vulnerability verification, detection and repair was kept, and data for offensive uses was removed.

How do I access Qwen4-Max, and what does it cost?

Talk to our team to plan access for your region. Availability, features and pricing vary by region and may change over time.

How is my data handled?

Your Alibaba Cloud team will walk you through the data processing, residency and retention terms that apply in your region before a pilot starts.

How do I get started?

Talk to Alibaba Cloud to scope a pilot: talk to our team.

Key terms

Work360. The Qwen industry benchmark framework covering 26 top-level domains, 182 sub-domains and 721 fine-grained domains of professional work.

Reasoning effort. A setting that trades depth of reasoning against tokens and cost. Qwen4-Max results are reported at High and Xhigh.

Live Delivery. Running a long task as a series of usable versions delivered as the work progresses, so people can review and steer it.

MCP tools. Tools connected to the model through the Model Context Protocol, such as market-data and search services.

Last Secure Code Bench. A secure-coding benchmark. Results here are the function-level rate of code that passes both functional and security checks.

In the case shown. A result from one specific launch demonstration. It is not a guarantee of results on other tasks.


Terms and disclaimers

This document is provided for general information only.

Results described as "in the case shown" come from specific Qwen4-Max launch demonstrations. Results on your own tasks depend on the task, tools, data and configuration, and will vary.

Benchmark and cost figures are as reported by the Qwen team on the named evaluations (Qwen Work360, Terminal-Bench 4.0 and Last Secure Code Bench) and compare Qwen4-Max with Qwen3.8-Max. They are not comparisons with any other provider's models.

Demonstration footage and images are from official Qwen4-Max demonstrations. Laboratory, manufacturing and research results reflect those demonstration settings.

Product and service availability, features, pricing and program terms vary by region and may change over time.

Nothing in this document constitutes professional, legal, medical or investment advice. Outputs in regulated work such as legal, financial and engineering tasks should be reviewed by qualified professionals.

Blender, Unreal Engine, Excel and PowerPoint are trademarks of their respective owners. Their mention does not imply endorsement.

© 2026 Qwen. All rights reserved.

0 0 0
Share on

Johnny Mai

22 posts | 2 followers

You may also like

Comments

Johnny Mai

22 posts | 2 followers

Related Products

  • Token Plan

    Build more, spend less. One plan, every modality.

    Learn More
  • Qwen

    Full-range, open-source, multimodal, and multi-functional

    Learn More
  • Alibaba Cloud Model Studio

    A one-stop generative AI platform to build intelligent applications that understand your business, based on Qwen model series such as Qwen-Max and other popular models

    Learn More
  • QwenWork

    QwenWork is dedicated to helping employees strengthen their professional competitiveness in the AI era and to enabling enterprises to improve organizational effectiveness.

    Learn More