SIAgentNews.com
← All news

Alibaba's QwenWork lets users pick rival Chinese AI models for office tasks

Alibaba's office agent now offers models from Kimi, GLM, DeepSeek, MiniMax and Step, and companies can set who sees which ones.

· First reported by pandaily.com

The short version

  • QwenWork added models from Kimi, GLM, DeepSeek, MiniMax and Step on November 16
  • Available to personal and enterprise standard-edition users
  • Alibaba recommends its new Qwen4-Max for multi-step complex reasoning
  • Admins can set model access by department, user group or individual
  • Alibaba gives no shared benchmark showing which model is best for each task

Why it matters

A company that sells an AI workspace is now letting competitors' models run inside it, which treats model choice as a routine setting rather than a lifelong commitment. For organizations, the permissions layer is the practical part: it decides which models staff can use for which work. For individuals, it is a cue to test more than one model on real tasks.

What Alibaba changed

Alibaba's QwenWork, an AI agent built for office work, widened its model menu on November 16. As pandaily.com reported, citing Alibaba's announcement as covered by Leiphone, users can now choose models from Kimi, GLM, DeepSeek, MiniMax and Step, in addition to Alibaba's own Qwen family.

Until now, people using the standard edition picked among Alibaba's own tiers: standard, advanced and Qwen3.8-Max. The new menu is open to personal users and to enterprise users on the standard edition.

This is a menu inside Alibaba's product. It is not a joint release, and the companies behind the added models did not co-launch anything with Alibaba.

Why offer a choice at all

Alibaba's reasoning is practical. As people give agents more specialized jobs, models differ in how well they reason, how quickly they answer and how many tokens they burn. A token is a small chunk of text that AI services count and usually charge for, so heavier use costs more.

The published menu includes Kimi-K3.1-Pro, several GLM and DeepSeek variants, MiniMax-M3.1-Flash and Step-5-Preview. Alibaba's newly released Qwen4-Max is available too. The "Flash" and similar labels generally point to lighter, quicker models, though the announcement does not define them.

Which model for which job

Alibaba offers guidance rather than a ranking. For work that needs many steps of difficult reasoning, it points to Qwen4-Max. The company says that model can produce a first usable draft of a result and then extend and revise it as the user gives feedback. It also describes the model as able to plan, work through a problem and check its output against demanding goals, with website design and cloud-resource scheduling as examples.

For analytical and presentation-heavy jobs, such as sorting out rules, checking plans, running simulations and building 3D slide decks, Alibaba suggests GLM-5.3 and DeepSeek-V4-Pro.

For quick, routine chores like drafting text, pulling details out of documents and tidying content, it leans toward cheaper, lighter models. It names Flash versions from Qwen, GLM and DeepSeek in that group.

These are the company's own claims. The announcement offers no independent comparison and no common benchmark showing which option wins at which task.

Control for administrators

The most concrete change for businesses is on the permissions side. According to Alibaba, enterprise administrators can set model access by department, by user group or for individual members.

That means a company need not show the same list to everyone. A legal team could be limited to one set of models, while a marketing group gets cheaper, faster options for everyday drafting. It also gives the organization a way to keep costs and risk in check, since the person choosing a model is no longer the only one deciding.

The update comes shortly after QwenWork 2.0, which added team workspaces. The new release changes the model-selection and access layer, not the underlying models themselves.

What the report leaves out

The source is silent on several things a buyer would want to know. It does not give pricing for individual models or say whether the menu is available outside China. It also does not explain where files and prompts go when a third-party model handles a task inside Alibaba's platform.

That last point deserves care. Data handling inside someone else's platform may differ from the model maker's own service, so anyone using this for sensitive work should ask before relying on it.

The release is best read as a sign of where office AI is heading: one workspace, many models, and a growing need for people to judge which one fits the job.

What to watch

  • Will Alibaba publish pricing or comparison data for the third-party models?
  • Is the expanded menu offered to users outside China?
  • How will QwenWork handle data when a rival model processes a task?

Try it

Take one real task, such as summarizing a long document, and run it through two or three different models in whatever platform you use. Compare quality, speed and cost, and check where your data goes.

Sources and links

First reported by pandaily.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news