Google has announced a preview of its Gemini 2.5 Computer Use model, part of Project Mariner, designed for AI-driven interactions with graphical user interfaces (GUIs) like browsers. This specialized model follows a looped process: it receives a user request alongside a screenshot and recent action history, analyzes these inputs, and generates a corresponding UI action, such as clicking or typing. In response, the client-side code executes the action, which restarts the cycle with updated screenshots and URLs.
The model supports various actions, including navigating URLs, scrolling, and organizing tasks. Google showcased its capabilities with examples involving pet care management and task organization for art clubs, emphasizing that the model excels in web and mobile UI tasks, outperforming competitors in browser control latency. Available for public preview via the Gemini API in Google AI Studio, Gemini 2.5 aims to enhance workflow automation for developers and speed up software development processes.
Source link
