Gemini 2.5: Google's new AI that uses the computer like a person

Google introduced Gemini 2.5, an improved version of its artificial intelligence model capable of interacting with digital interfaces like a human. Thanks to its visual reasoning and iterative execution, the system can operate platforms, complete forms, and organize tasks autonomously.

The model is already available for developers in public preview through the Gemini API in Google AI Studio and Vertex AI.

Google presentó la versión final de sus modelos de inteligencia artificial Gemini 2.5 Pro y Flash

A model that acts and reasons like a human user

Unlike traditional systems, which rely on structured APIs, Gemini 2.5 can manipulate graphical interfaces directly. This includes typing, clicking, scrolling, using dropdown menus, or navigating between pages, even within platforms that require login.

Completes and submits online forms.
Navigates websites or collaborative platforms.
Sorts, moves, and organizes items according to user instructions.

For example, the system can arrange notes on a digital task board by following precise instructions.

How the Gemini 2.5 model works

The model operates through the computer_use tool included in the Gemini API. It works in an iterative cycle: the user sends a request along with a screenshot and the history of recent actions. Gemini analyzes that data, generates a response, and executes an action (such as clicking or typing).

Google Gemini es una herramienta poderosa, pero no está diseñada para manejar ciertos tipos de datos

After each action, the system receives a new screenshot of the environment and repeats the process until the task is completed or a stop command is received. This method enables safe and controlled interactions with the digital environment.

Platform performance and security

According to Google, the model outperformed leading alternatives in web and mobile control tests, showing lower latency and higher accuracy. Although it is optimized for browsers, it also showed promising results in other interfaces.

Seguridad y confiabilidad en la información

In terms of security, Google implemented multiple layers of protection. Among them, a step-by-step verification system that evaluates each action before executing it, such as purchases or access to personal data.

Availability of Google's new model

Since October 7, Gemini 2.5 has been available for testing in Google AI Studio and Vertex AI. Additionally, developers can experiment with the model in a demonstration environment hosted by Browserbase or create their own environment with tools such as Playwright.