Google Launches Gemini 2.5 Computer Use: AI That Browses the Web Like a Human

Google has unveiled Gemini 2.5 Computer Use, a specialized AI model built on Gemini 2.5 Pro, designed to interact with the web like a human. Unlike traditional AI, which generates text or answers questions, this model physically performs actions on websites — such as:

Filling and submitting forms

Clicking buttons and links

Typing text

Scrolling pages

Using keyboard shortcuts

Hovering over elements

Opening dropdown menus

Currently, it supports 13 distinct actions, and Google says it outperforms competitors in web and mobile benchmark tests while maintaining lower latency.

Demo in Action

Google showcased demo videos (sped up 3x) where Gemini 2.5 Computer Use organizes tasks on a web app. For example, it can:

“Go to stick-note-jam.web.app and ensure notes are placed in the correct category, dragging them if necessary.”

This demonstrates the model’s ability to understand instructions and interact dynamically with web pages.

Current Availability and Limitations

Available to developers via: Google AI Studio and Vertex AI

Supported actions: 13 web-based actions

Limitations: Can only access browser-level tasks, not optimized for desktop OS-level control yet

Applications and Use Cases

Google teams are already using Gemini 2.5 Computer Use for:

UI testing — speeding up software and web interface testing

Agentic AI applications in AI Mode in Search, Firebase Testing Agent, and Project Mariner

Assigning AI agents to handle tasks like research, planning, and data entry

The model is a step toward AI that can execute complex workflows across websites automatically, making digital workflows more efficient for developers and businesses.