Google is previewing a caller Gemini AI exemplary designed to navigate and interact pinch nan web via a browser, letting AI agents do things wrong interfaces designed for usage by group and not robots. The model, called Gemini 2.5 Computer Use, uses “visual knowing and reasoning capabilities” to analyse a user’s petition and transportation retired a task, specified arsenic filling retired and submitting a form.
It tin beryllium utilized for UI testing aliases navigating interfaces made for group who don’t person an API aliases different nonstop relationship available. Other versions of this exemplary person been utilized for agentic features successful AI Mode and Project Mariner, a investigation prototype that uses AI agents to transportation retired tasks connected its ain successful a browser, for illustration adding items to your cart based connected a database of ingredients.
Google’s announcement comes conscionable 1 time aft OpenAI revealed new apps for ChatGPT arsenic portion of its yearly Dev Day, and continues to attraction its attraction on its ChatGPT Agent feature that tin complete analyzable tasks connected your behalf. Meanwhile, Anthropic had already released a type of its Claude AI exemplary pinch “computer use” past year.
Google posted immoderate demo videos showing its machine usage instrumentality successful action, and notes that they are sped up 3x.
Google says its machine usage exemplary “outperforms starring alternatives connected aggregate web and mobile benchmarks.” Unlike ChatGPT Agent and Anthropic’s machine usage tool, Google’s caller AI exemplary only has entree to a browser — not an full machine environment. Google notes that it shows “it is not yet optimized for desktop OS-level control” and presently supports 13 actions, including opening a web browser, typing text, arsenic good arsenic dragging and dropping elements.
Gemini 2.5 Computer Use is disposable to developers done Google AI Studio and Vertex AI, but there’s besides a demo connected Browserbase, wherever you watch arsenic it completes tasks, for illustration “Play a crippled of 2048” aliases “Browse Hacker News for trending debates.”
3 months ago
English (US) ·
Indonesian (ID) ·