Google’s latest AI model uses a web browser like you do

 

What is Gemini 2.5 Computer Use?

Gemini 2.5 Computer Utilize is a form of Google’s Gemini AI that is particularly adjusted to control a web browser interface to carry out assignments. 

VentureBeat

+2

Silicon ANGLE

+2

It employments visual understanding and thinking over client interface formats (buttons, content boxes, dropdowns, etc.). 

The Verge

+2

Silicon ANGLE

+2

It bolsters a settled set of interface activities (such as “open browser,” “click,” “type,” “drag”) — as of now detailed to be 13 activities. 

The Verge

+2

VentureBeat

+2

The show works in a circle: it gets the current UI (screenshot, state, activity history), reasons around what to do following, issues an activity call, at that point gets the coming about unused UI state, and proceeds until the errand is done. 

Silicon ANGLE

+2

BGR

+2

Because it is restricted to browser intuitive, it cannot control the computer exterior of the browser (e.g. getting to nearby records, OS-level control). 

The Verge

+1

Why this is noteworthy / what it enables

Many web applications (keeping money locales, shapes, dashboards, etc.) don’t give APIs or clean endpoints for outside mechanization. An AI that can utilize the UI itself gets to be valuable. 

VentureBeat

+2

Silicon ANGLE

+2

It decreases the crevice between what a human can do in a browser and what robotized instruments can do. The AI can “see” the page structure and choose how to associated. 

The Verge

+1

It proposes a future where AI colleagues might do more “hands-on” work online (booking tickets, overseeing shapes, shopping) or maybe than as it were creating content or recommendations. 

The Verge

+1

Because it’s restricted to browser-level interaction (not full framework control), Google can more firmly control what the AI is permitted to do, likely decreasing security and security dangers. 

The Verge

+1

How this compares with what I (ChatGPT) can do

Right presently, I don’t have built-in capability to specifically control a browser UI (tap, sort, scroll) in a live web environment. My part is to produce content, direction, plans, code, or instructions.

When I “browse” (by means of instruments), I utilize look APIs and recover substance, or maybe than visual UI control.

Google’s unused demonstrate is more “agentic” in the sense of taking activities on a UI, though my intelligent are more unique (you tell me what you need, I react with text/instructions or get data).