Alibaba adds Kimi, GLM, and DeepSeek models to QwenWork office agent
Alibaba is opening its office agent to third-party AI models, allowing users to choose specific tools based on speed, cost, and reasoning needs.

The short version
- QwenWork now supports models from Kimi, GLM, DeepSeek, MiniMax, and Step in addition to Alibaba's own Qwen series.
- The update allows users to select models based on specific needs like reasoning ability, response speed, and token consumption.
- Enterprise administrators can control which models are visible to different departments or individual users.
- Alibaba recommends specific models for different tasks, such as Qwen4-Max for complex reasoning and Flash models for quick drafting.
Why it matters
This shift moves the AI workspace from a single-provider model to a multi-model ecosystem. It gives businesses more flexibility to optimize for cost and performance depending on the specific job at hand.
A multi-model office agent
Alibaba recently updated its QwenWork office agent to include a much broader selection of AI models. Starting November 16, the platform began offering third-party options from several different providers. These include models from Kimi, GLM, DeepSeek, MiniMax, and Step, which now sit alongside Alibaba's own Qwen lineup.
This change is available to both personal users and those on enterprise standard editions. By expanding the menu, Alibaba is moving away from a system where users were restricted to its own standard and advanced options. Instead, the platform now functions as a hub where different models can be swapped in depending on the specific requirements of a task.
The expansion is not a joint release between Alibaba and these other companies. Rather, it is a functional update to the QwenWork interface. It allows the office agent to draw from a wider set of providers while the user or the organization decides which specific model is best suited for a particular job.
Choosing by task and performance
Alibaba stated that this expansion responds to the fact that different tasks require different types of AI performance. For example, some jobs might require high reasoning abilities, while others might prioritize response speed or lower token consumption. By offering a variety of models, the platform allows users to tailor the agent's behavior to the specific work they are handing off to the AI.
The available menu includes high-end options like Kimi-K3.1-Pro, GLM-5.3, and DeepSeek-V4-Pro. Alibaba specifically suggests using its newly released Qwen4-Max for work that involves multiple steps of complex reasoning. The company claims this model can provide an initial usable version of an output and then revise it based on user feedback.
For everyday tasks that need to be done quickly, Alibaba points users toward lower-priced
Specialized task recommendations
Alibaba provides specific guidance on which models to use for different types of professional projects. For instance, the company suggests using GLM-5.3 and DeepSeek-V4-Pro for tasks that involve organizing rules, reviewing plans, running simulations, or building 3D presentations. These models are positioned as the go-to choices for these more structured and specialized workflows.
For high-speed daily operations, the platform recommends using lower-priced options. These include Qwen3.8-Flash, GLM-5.3-Flash, and DeepSeek-V4.1-Flash. These models are intended for faster work such as drafting copy, extracting specific information from documents, and organizing general content.
Complex goals and constraints
The company describes the capabilities of its high-end models in terms of planning, solving, and checking work against complex goals and constraints. Alibaba highlights specific use cases for these abilities, such as website design and cloud-resource scheduling. These examples illustrate how the agent can handle multi-layered instructions rather than just simple prompts.
While Alibaba makes these claims about the capabilities of its models, it is important to note that these are the company's own descriptions. The announcement does not provide any independent comparison or a common benchmark to establish which specific model is the best option for every possible task.
Enterprise control and new features
The update allows enterprise administrators to manage how these models are distributed across a company. Administrators can configure model access by department, user group, or even individual members. This means an organization can offer different choices to different roles instead of exposing the entire menu to every employee, allowing for more precise control over tool usage.
This model-selection update follows the recent release of QwenWork 2.0, which introduced team workspaces and event-triggered tasks. By combining these new workspace features with a wider variety of third-party models, Alibaba is creating a more flexible environment for teams to manage specialized AI workflows.
Sources and links
- Reportingpandaily.com: the article we sourced
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.