Search

Alibaba Cloud launches an AI built to run long, multi-tool enterprise workflows

Alibaba Cloud unveiled Qwen4-Max, a 2.4-trillion-parameter enterprise model with native image understanding and a 1,000,000-token context window designed to run extended, tool-heavy workflows...

Server racks and network cables inside a data centre.
Photo by Taylor Vick on Unsplash

Alibaba Cloud has announced Qwen4-Max, a 2.4-trillion-parameter enterprise model the company describes as an "always-on AI colleague". Alibaba says the model reads images natively and holds up to a 1,000,000-token context window - the amount of text the model can keep in working memory while it completes long tasks - so it can run extended, tool-heavy workflows and produce finished outputs rather than single answers.

Alibaba lists technical details including 2.4 trillion total parameters (the model's internal settings and a rough measure of its size), about 71 billion active parameters per token (the company's measure of how much of the model is used at each step), 328 billion N-gram embedding parameters, and a new Gated DeltaNet architecture with Qwen Sparse Attention. The announcement highlights a Live Delivery capability, which Alibaba describes as a way to ship an early usable version of a long task and then deliver improvements as the work continues. The company positions these features as ways to keep very long contexts efficient and to integrate tool use into extended runs.

Alibaba says the results it showed come from scoped demonstrations and named benchmarks and are not guarantees for other users or workflows. In those launch demonstrations Alibaba reported time and cost improvements in specific cases, for example:

  • Manufacturing: Alibaba says Qwen4-Max generated and simulated CAM toolpaths and, in the demonstration shown, reduced hands-on CAM programming time per part from about 475 minutes to 24 minutes.
  • Science: Alibaba says that in a 48-hour autonomous run combining computation with a robotic wet lab, the model explored a 2.7-million-molecule candidate space and produced a coolant molecule validated in the lab demonstration.
  • Legal: Alibaba says for an IPO due diligence demonstration Qwen4-Max organised 198 unstructured files and produced an 86-page cited report covering multiple legal issues.
  • Finance: Alibaba says the model, using many connected tools, assembled market data, cross-checked sources and produced workpapers and an editable slide deck with source tracing.
  • Cloud operations: Alibaba reports that 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 tests more than 14.61 times faster.

Alibaba reports Qwen4-Max scored 52.69 on its Qwen Work360 evaluation versus 47.74 for Qwen3.8-Max, that it used fewer tokens at higher reasoning settings, and that average cost per Work360 task was about 58% lower on the Qwen Work360 evaluation. For security-related metrics, Alibaba cites a rise in function-level secure and functional code on the Last Secure Code Bench, from 22.0% to 31.8%.

On infrastructure, Alibaba Cloud's global locations page lists 106 availability zones in 31 regions. Alibaba's pages also say customers should select the data centre closest to their end users to speed access, and that "your Alibaba Cloud team will walk you through the data processing, residency and retention terms that apply in your region" before a pilot starts. Availability, features and pricing vary by region, Alibaba says.

Because Alibaba does not list a public African region on its published regions page, organisations in Africa that want to use Alibaba Cloud would typically host workloads in regions outside the continent. That can mean higher latency for users in Africa compared with running in a local data centre, and it can require cross-border data transfers - matters that may affect performance and data-residency or compliance arrangements depending on local law and the customer's needs.

For comparison, Amazon Web Services, Microsoft Azure and Google Cloud each publish pages listing their public regions: https://aws.amazon.com/about-aws/global-infrastructure/, https://azure.microsoft.com/en-us/global-infrastructure/regions/, https://cloud.google.com/about/locations

Will Qwen4-Max be available through local partners or a dedicated African region, and what data-residency options will Alibaba offer for African customers?

Newzroom dot Africa has reached out to Alibaba Cloud for comment and is awaiting a response. This story will be updated when they respond.

Source: Alibaba Cloud

Salute the work behind this story.

Blurbs

Up to 500 characters. Your name and location show with your blurb; your email is never shown. No links, email addresses or phone numbers in the blurb. Blurbs that break the terms of use are removed.

    More like this