Alibaba Introduces Qwen4-Max for Work, Creative Production and Research

What Readers Should Know
Alibaba has announced Qwen4-Max for multimodal work and complex agent workflows. Its efficiency figures come from company evaluations, while smaller open-weight Qwen4 models are planned separately.
Key Takeaways
- Qwen4-Max supports multimodal tasks.
- Alibaba reports a one-million-token context window.
- Its architecture activates a subset of total parameters.
- Benchmark savings are company-reported.
- Smaller open-weight models are described as forthcoming.
Alibaba has announced Qwen4-Max, a new multimodal AI model designed for complex work, creative tasks, and research workflows.
In its November 18 release, the company describes a model with native vision capabilities, a one-million-token context window, and 2.4 trillion total parameters.
Alibaba’s official documentation identifies programming, office work, front-end design, and complex agent tasks among its intended uses.
A large model with selective activation
According to Alibaba, Qwen4-Max uses a Mixture-of-Experts architecture, activating 71 billion parameters per token rather than using its entire parameter count for each processing step.
The company presents this design as a way to improve capability while reducing the computing cost of generating responses.
Its release reports efficiency gains over Qwen3.8-Max on Work360, Alibaba’s proprietary benchmark covering tasks across 721 specialized fields.
Alibaba says the model used 35% fewer tokens in high-effort reasoning mode and reduced average cost per task by around 60% in that evaluation. These are company-reported benchmark results, rather than guaranteed savings for every customer.
From coding to longer-running workflows
The announcement describes applications in 3D modeling, game development, video creation, cloud scheduling, and manufacturing.
It also reports scientific demonstrations involving extended agent workflows. These examples should be understood as research and showcase results—not evidence of an approved medical treatment or universally autonomous scientific capability.
Access and future releases
Alibaba says Qwen4-Max is available through QwenCloud, including its Token Plan and API, alongside QwenWork and QwenStudio.
The release quotes US$0.075 per million input tokens for implicit cache hits. That is a specific billing category, not the complete price of every request.
Smaller open-weight Qwen4 models are described as forthcoming. Their planned release should be distinguished from Qwen4-Max’s current access arrangements.
Frequently Asked Questions
What is Qwen4-Max?
Alibaba’s newly announced multimodal AI model.
What tasks does it target?
Coding, office work, creative production, and complex agent workflows.
Are the reported savings guaranteed?
No. They reflect the company’s stated evaluation conditions.
Is US$0.075 the complete request price?
No. The release identifies it as an implicit-cache-hit input rate.
Are smaller open-weight Qwen4 models already released?
The supplied announcement describes them as forthcoming.