Google releases Gemini 2.5 Computer Use in public preview

Google on Thursday made the Gemini 2.5 Computer Use model available in public preview to developers, unveiling a specialized artificial intelligence that can operate graphical user interfaces in web browsers and, to a lesser extent, mobile apps. The model, built on Gemini 2.5 Pro, accepts natural-language instructions and then performs browser actions such as clicking, typing, scrolling, dragging and dropping, filling and submitting forms, and navigating pages to complete tasks.

The system operates in an iterative loop. A user request is submitted alongside a screenshot of the current interface and a short history of recent actions. The model analyzes those inputs and returns a function call that represents a specific UI action, which client-side code executes. After the action runs, the updated screenshot and URL are returned to the model and the cycle repeats until the task is finished, an error occurs, or a safety rule or user decision stops the interaction. Google said the computer_use tool in the Gemini API exposes these capabilities, and developers can access the model through Google AI Studio and Vertex AI.

Google said the model is primarily optimized for web browsers but...