Gemini 2.5 Computer Use Model Officially Introduced: Now Available as Preview Via Gemini API in Google AI Studio and Vertex AI

Google has now officially introduced its Gemini 2.5 Computer Use model, and it has been brought out as a specialized model built on the capabilities of the Gemini 2.5 Pro model. Its preview version is now available via the Gemini API in Google AI Studio and Vertex AI.

Check out more about it below.

Google’s Newly-Introduced Gemini 2.5 Computer Use Model Now Available as a Preview Version

As announced earlier this year in 2025, Google has introduced its Gemini 2.5 Computer Use model and via the Gemini API has brought computer use capabilities to developers. The brand’s new model has been built over the visual understanding and reasoning capabilities of the Gemini 2.5 Pro model, which can be used to power agents that are capable of interacting with UIs (user interfaces). Featuring comparatively lower latency, the model outperforms other alternatives on both web and mobile based benchmarks.

Speaking of how it works, relying on the new ‘computer_use’ tool in the Gemini API, and inputs which include the ‘user request’, ‘screenshot of environment’, and ‘history of recent actions’, the model exhibits its capabilities. Additional custom functions can also be specified if needed. Then, analyzing the inputs, a response will be generated. An end user confirmation may also be a part of the response (such as for payments). Afterwards, the received action will be executed by the client side code. Once the action gets executed, the new GUI screenshot and the current URL are sent again to the model, and these will serve as a function response that restarts the loop. The process then continues until the given task is completed.

Notably, the Gemini 2.5 Computer Use model has been optimized for web browsers, and it is additionally capable of handling mobile UI control tasks as well. However, as of now, it has not yet been optimized for desktop OS-level control. On the safety side, the necessary training has been given for the model, and key risks which include ‘intentional misuse by users’, ‘unexpected model behavior’, and ‘prompt injections and scams in the web environment’ have been addressed, providing safety controls like ‘per-step safety service’ and ‘system instructions’.

As mentioned, from today (8th October 2025), the Gemini 2.5 Computer Use model is available to developers as a public preview version, and it can be accessed via the Gemini API in Google AI Studio and Vertex AI.