AI library

Source · docs

Gemini API Computer Use generally available

Author
Google
Published
30 October 2026
Added
2 November 2026
Original
https://aistudio.google.com/docs/computer-use

Summary

The Gemini API changelog for 2026-10-30 says the Computer Use tool is now generally available in both the Interactions API and the GenerateContent API, with no other details. The Computer Use docs page describes the tool: the model proposes browser, desktop, or mobile actions and your own client carries them out.

Key points

  • Models: Gemini 4 Argon, Gemini 3.8 Flash (the page's recommendation for computer use), Gemini 3.5 Flash-Lite (lower latency and cost), and Gemini 3 Flash Preview.
  • Enable it: add a computer_use tool with an environment (browser, desktop, or mobile). Python needs google-genai 2.7.0 or later. Coordinates come on a 0–999 grid, which your client scales to the screen.
  • Actions: the browser has the widest set (navigate, go_back, scroll, drag_and_drop, hotkey); mobile adds open_app, list_apps, long_press; desktop uses OS-level mouse and keyboard. You can exclude predefined functions.
  • Safety: each call may carry a safety_decision (for example require_confirmation); you must ask the user and acknowledge it in the function result. Policies cover payments, sensitive data, communications, account creation, and legal terms. Prompt injection detection is opt-in (enable_prompt_injection_detection) and needs 3.5 Flash-Lite or later.
  • Google still advises close supervision, sandboxed environments, and allowlists.
  • The docs page itself does not mention GA status or prices (checked 2026-11-02).

Why it matters

Before this, computer use on Gemini models was a preview feature. GA on Flash and Flash-Lite makes cheap, client-run browser and desktop automation an option alongside Claude's computer use and OpenAI's hosted computer use.