Google's Gemini API release notes for October 30 list one change: the Computer Use tool is generally available across the Interactions and GenerateContent APIs. It had been in preview since June, when it launched on Gemini 3.5 Flash.
The model doesn't run anything itself. You send a prompt and a screenshot, it returns an action such as click, type, scroll or navigate with x and y on a 0-999 grid, and your code runs that action (Google's examples use Playwright) and sends back a new screenshot. Each action carries an intent field that says why the model chose it. There are three environments: browser, mobile (Android-style actions like open_app and long_press) and desktop.
The docs name gemini-3.8-flash as the recommended model, with gemini-4-argon for harder work and gemini-3.5-flash-lite as the cheap option. A built-in safety layer can mark an action as require_confirmation, which your app has to show to a person before it continues. You can turn off some policy categories, such as data_modification, but Google says the model may still ask for confirmation. Prompt injection detection, which scans screenshots for hidden instructions, is off unless you set enable_prompt_injection_detection to true.