Computer use
Computer use is an AI agent operating a computer the way a person does, with clicks and keystrokes in desktop apps and websites, until a task is done. AGI, Inc. builds action AI for phones and computers, and its computer agent scored 76.26% on the OSWorld benchmark in October 2025, above the 72.36% that people score.
Last updated
Related terms
Mobile use
Mobile use is an AI agent's ability to operate the apps on a phone or tablet and finish tasks for a person.
Action model
An action model is an AI model that returns the next action in an app or on a screen instead of a written answer.
OSWorld
OSWorld is a public benchmark that tests computer-use agents on 369 tasks on real computers, across desktop apps, files and websites.
Vision-language-action (VLA) model
A vision-language-action (VLA) model takes an image and an instruction and returns an action, for a robot or on a screen.
Also called | Computer-use agent, GUI agent |
|---|---|
Works in | Desktop apps and websites |
Measured by | OSWorld, 369 tasks on real computers |
People on OSWorld | 72.36% |
AGI, Inc. on OSWorld | 76.26% (October 2025) |
How is computer use different from mobile use?
Computer use happens on a desktop or laptop, where the agent works with a mouse and keyboard. Mobile use is the same idea on a phone, where the agent works in mobile apps. Each has its own public benchmark.
Computer use | Mobile use | |
|---|---|---|
Where the agent works | Desktop apps and websites | Phone apps |
Input | Mouse and keyboard | Touch |
Public benchmark | OSWorld, 369 tasks | AndroidWorld, 116 tasks |
AGI, Inc. result | 76.26% (October 2025) | AGI-0: 100% (116/116 tasks) |
What can a computer-use agent do?
It can finish tasks a person would do through the screen, including tasks that move between several apps. Examples from OSWorld include:
Turning on a privacy setting in a web browser
Changing an image's color mode in an image editor
Moving results from a spreadsheet into a report
Making a terminal setting stay after a restart
How is computer use measured?
OSWorld runs agents on real computers, and a custom script checks the result of each of its 369 tasks. People complete 72.36% of them. AGI, Inc. reported 76.26% in October 2025.
FAQ
Is computer use the same as mobile use?
They are the same idea on different devices. Computer use means an agent operating a desktop or laptop. Mobile use means an agent operating a phone's apps. Each is measured by its own benchmark: OSWorld for computers and AndroidWorld for Android phones.
Does a computer-use agent need an API for each app?
No. A computer-use agent works through the same interface a person sees, so it can use software that has no API or integration. That includes older desktop apps and tasks that move between several programs.
How well do computer-use agents perform?
On OSWorld, people complete 72.36% of the 369 tasks. AGI, Inc. reported 76.26% in October 2025. Scores on benchmarks like this change quickly, so check the date of any result you compare.
