Xiaomi Becomes First China-Based Lab To Offer Computer Use, Unveils MiMo Desktop Agent In Invite-Only Beta
Chinese know-how firm Xiaomi introduced that it has opened invitation-only testing for MiMo Desktop, a desktop AI software that executes full workflows fairly than producing responses in a chat interface. The launch positions the corporate as the primary China-based laboratory to supply computer-use capabilities to customers by way of its flagship fashions.
Real-world work not often begins with a well-defined immediate. It sometimes includes spreadsheets, pictures, video, PDF paperwork, audio recordings, and compressed information that collectively represent the context of a process. MiMo Desktop accepts these multi-format inputs immediately, with out requiring prior group or format conversion. Users state their goals in pure language, after which the system interprets the supplies, decomposes the duty, invokes applicable instruments, and delivers editable outputs — together with paperwork, spreadsheets, shows, net pages, audio, video, 3D fashions, and software program engineering initiatives.
Two design choices distinguish the product. First, the outcome preview isn’t a static rendering however a completely interactive deliverable incorporating element construction, interplay logic, and information visualization, which customers can function and revise throughout the session — whether or not the output is a knowledge dashboard, a recreation prototype, or a presentation. Second, modifications are made by choice: customers spotlight a chart, paragraph, or area and describe the specified change, and solely that portion is regenerated. Every revision is preserved in model historical past, permitting customers to check drafts and roll again to earlier states.
The system additionally incorporates “Smart” scheduling. Rather than requiring customers to pick out a mannequin manually, MiMo Desktop assesses process kind, complexity, and price necessities, routing routine work to straightforward fashions for pace and sophisticated, multi-step duties to flagship fashions for larger output high quality. Large-scale duties could be distributed throughout a number of agent classes that collaborate whereas sustaining unbiased reminiscence and workspaces. Xiaomi stories that cache optimization achieves hit charges of as much as 99% inside a session on lengthy duties, which reduces redundant computation and controls operational prices.
Browser and Computer Control as Core Capabilities
The second dimension of the discharge issues management. MiMo Desktop operates the browser as each an data supply and an execution surroundings: it opens pages, retrieves data, completes varieties, extracts property, and imports findings immediately into the duty. When producing web-based outputs, the system can moreover confirm deliverables by way of the browser throughout the workflow.
The most necessary function is full pc management, obtainable within the abroad model. MiMo Desktop reads display content material and operates the keyboard and mouse throughout purposes — opening information, verifying information, and transferring data between applications — with built-in outcome verification that may set off corrections or halt execution when needed. For steady and repetitive processes, a Record & Replay operate permits customers to exhibit a workflow as soon as and have the system re-execute it by way of pure language directions.
This displays a deliberate architectural strategy: the desktop is handled because the working surroundings, and the mannequin features as an agent inside it, fairly than as a language interface layered on high of current purposes.
A Year within the Making: How Xiaomi Built the Technical Case for Computer Use
Computer-use functionality isn’t primarily an interface downside however a technical one, and Xiaomi has been growing the underlying basis for greater than a yr. The groundwork was established in mid-2025 with MiMo-VL-7B, the corporate’s open-source vision-language mannequin, which achieved a then-record rating of 56.1 on OSWorld-G — the usual benchmark for GUI grounding — outperforming fashions developed particularly for the duty, comparable to UI-TARS. This functionality was developed by way of a four-stage coaching pipeline protecting cellular, net, and desktop interfaces, along with a considerable corpus of Chinese GUI information and a coaching process that infers intermediate actions from before-and-after screenshots — exactly the perceptual basis required for a computer-use agent.
The mannequin lineage has continued to develop. By early 2026, Xiaomi’s flagship MiMo-V2-Pro ranked in the identical tier as Claude 4.5 Sonnet, GPT-5.2, and Gemini 3.0 Pro on coding agent, normal agent, and tool-use benchmarks. This gives the Desktop shopper with a top-tier mannequin for complicated duties whereas the routing layer maintains price effectivity for routine operations. In addition, Xiaomi brings greater than 20 years of expertise in shopper {hardware} and software program ecosystems, which is mirrored within the product’s concentrate on the sensible circumstances of a consumer’s desktop surroundings fairly than on benchmark efficiency alone.
For observers of the AI sector, the discharge signifies a broader shift in analysis standards: the frontier is shifting from reasoning benchmarks towards operational competence — brokers that full varieties, confirm their very own output, and stay inside outlined compute budgets. Xiaomi’s mixture of GUI-grounding fashions, a frontier-tier flagship, and a computer-use interface now in beta locations the corporate among the many main contenders within the rising class of autonomous desktop brokers.
The submit Xiaomi Becomes First China-Based Lab To Offer Computer Use, Unveils MiMo Desktop Agent In Invite-Only Beta appeared first on Metaverse Post.
