A tool for an application you build
Google’s October 30 release notes make Computer Use generally available across the Interactions and GenerateContent APIs. [1]
The tool uses screenshots to propose interface actions for browser, mobile, or desktop environments. Your application supplies the execution loop. Google recommends gemini-3.8-flash; check the guide’s supported-model list. Prompt-injection screenshot scanning is opt-in. [2]
Confirmation and billing still matter
When a safety decision requires confirmation, the application must ask the user and acknowledge an approved action in its result. Configured overrides do not remove the need to handle those decisions. [2]
The pricing page lists Computer Use as unavailable on the Free tier. On the Paid tier it is charged as regular tokens at the selected model’s price. [3]
Our take
For technicians, treat the action handler as part of the product you must review. Choose permissions for the task, explain consequential actions, and verify the resulting changes. Start with synthetic data and a clear stopping condition. We reviewed documentation without calling the API.