Google recently previewed a new Gemini AI model - Gemini2.5Computer Use, designed to give AI agents the ability to navigate and interact with web pages through a browser. This model uses its powerful "visual understanding and reasoning capabilities", enabling it to analyze user requests like humans and perform complex operations within interfaces originally designed for humans, not robots, such as filing out and submitting forms.
New Frontiers for AI Agents
Gemini2.5Computer Use enables AI to perform tasks that previously required human intervention. Its main application scenarios include UI testing, and navigating web interfaces for users who do not have an API or direct connection. The early version of this model was used in the Mariner project - a research prototype that used AI agents to complete tasks on their own in the browser, such as adding items to a shopping cart based on a list of ingredients.
The release of this new model comes at a time when competition for AI agent features is heating up. Just a day before Google announced it, OpenAI released a new ChatGPT app at its developer day and continues to focus on its Agent feature, which can complete complex tasks for users. At the same time, Anthropic also released a version of the Claude AI model with a "computer use" feature last year.
Performance and Limitations
Google claims that its Gemini2.5Computer Use model "outperforms leading alternatives in multiple web and mobile benchmark tests."
However, unlike ChatGPT Agent and similar tools from Anthropic, Google's new AI model currently can only access the browser environment, not the entire computer environment. Google points out that the model "has not been optimized for control at the desktop operating system level," and currently supports 13 operations, including opening a web browser, entering text, and dragging and dropping elements.
How to Experience It
Developers can now experience Gemini2.5Computer Use through Google AI Studio and Vertex AI.
For regular users and interested parties, Browserbase offers a demonstration
