Gemini API Computer Use is generally available

Computer Use on the Gemini API is generally available. The October 30, 2026 release notes say the tool was released to general availability across the Interactions API and the GenerateContent API.

A status change, not a new feature list

That entry is one line. It does not add environments, intents, safety controls, pricing, or benchmarks. Browser, mobile, and desktop support, the intent field, and safety settings are documented on the current Computer Use guide as the tool developers can call now. They are not presented there as announcements dated October 30.

The guide says the tool lets you build agents that control a browser, a mobile environment, or a desktop. The model "sees" the screen from screenshots and "acts" by returning UI actions such as mouse clicks and keyboard inputs. You still implement the client that receives and executes those actions. The API does not drive the interface itself.

What the current guide describes

For Gemini 4 and Gemini 3.x, the guide lists multi-environment support, an intent field that explains why the model chose each step, configurable safety policies with built-in categories and overrides, and opt-in screenshot scanning for hidden adversarial instructions. A response can also carry a safety_decision: regular/allowed, require_confirmation (the user must approve), or blocked. If the action is blocked, the client should stop or handle the interruption.

The documented loop is continuous. Send the Computer Use tool, the target environment, the prompt, and a screenshot. Read the function call. Scale coordinates from the model's 0–1000 grid to the viewport, run the action, capture a new screenshot, and return it as a function result until the task finishes or stops. Examples on the page use gemini-3.8-flash and show the Interactions API as the recommended surface, with a generateContent path documented alongside it. The guide's Model versions section is where supported model strings are listed; the October 30 note does not repeat that list.

Uses called out in the guide include repetitive data entry and form filling, automated testing of web applications and user flows, and research across sites such as gathering product information, prices, and reviews. The setup notes say to run the agent in a sandboxed VM or container and to write the client-side handler for clicks, typing, and screenshots.

Read the Computer Use guide before you wire that loop, and use the October 30 release note to separate general availability—on both Interactions and GenerateContent—from the tool surface the guide describes.

Sources

• Gemini API release notes
• Computer use