Google launches Gemini 2.5 Computer Use AI that can browse the web like humans

In a blog post, the company said the specialized model, built on Gemini 2.5 Pro’s visual understanding and reasoning capabilities, enables AI agents to interact directly with user interfaces (UIs) by clicking, typing, and scrolling.

“Today, we are releasing the Gemini 2.5 Computer Use model, our new specialized model built on Gemini 2.5 Pro’s visual understanding and reasoning capabilities that powers agents capable of interacting with user interfaces,” Google said.

According to the company, the model outperforms leading alternatives on several web and mobile control benchmarks while offering lower latency. Developers can access the new features through the Gemini API in Google AI Studio and Vertex AI.

While traditional AI systems rely on structured APIs to interface with software, Google noted that many real-world digital tasks still require direct interaction with graphical user interfaces — such as filling forms, submitting data, or navigating websites.

How it works

The Computer Use capability is integrated as a new tool within the Gemini API and operates in an iterative loop. It processes three main inputs — the user’s request, a screenshot of the environment, and a history of recent actions.

The model analyzes these inputs and generates a response, typically a function call representing a UI action like clicking or typing. Some actions, such as making a purchase, may prompt the model to request user confirmation before execution.

Once the action is executed, an updated screenshot and current URL are sent back to the model, continuing the interaction loop until the task is completed, an error occurs, or the session ends due to a safety response or user termination, Google explained.

Google said the Gemini 2.5 Computer Use model is currently optimized for web browsers, though it also shows strong potential for mobile UI control tasks. It is not yet tuned for desktop operating system-level control, the company added.