Google Gemini Computer Use allows the flagship AI model to execute multi-step commands directly across device interfaces.
UCG/Universal Images Group via Getty Images
AI is no longer trapped inside your browser tab or your voice assistant. A native Google Gemini Computer Use upgrade now gives Gemini 3.5 Flash the power to control your computer or mobile device directly, just as though it were moving your mouse or tapping your screen. It’s available now to developers, but here’s why you really shouldn’t enable it unless you know what you’re doing.
As revealed in a recent post on The Keyword blog, Google has just announced a radical new “Computer Use” feature for its Gemini 3.5 Flash AI model. This will allow the model to take over your device and perform actions on your behalf.
Google already offers consumers several agentic AI features, such as the ability to control a remote virtual computer and browser with Gemini Spark. But Gemini 3.5 Flash computer use is different: it controls the physical device in front of you.
Google previously offered this functionality as a separate Gemini 2.5 Computer Use model. By baking it directly into Gemini 3.5 Flash, developers can now invoke device control alongside standard capabilities like Search and Maps without switching to a dedicated model."
This upgrade addresses key limitations of the previous model, which was optimized primarily for browser-based control. According to Google, it should result in more responsive execution for “long-horizon and enterprise automation tasks.”
Google's Gemini 3.5 Flash agent autonomously navigating a mobile application interface to categorize features and execute multi-step tasks.
The prospect of an AI model taking full control of your device is, quite understandably, rather frightening. But to get past that initial “Hell no!” reaction, Google has implemented several safeguards.
The first, and most obvious, is that Google isn’t unleashing this capability on general consumers. Computer Use is aimed squarely at developers and enterprise environments with a need to automate tasks like testing new user interfaces, conducting research across various websites and apps, or automating data entry into legacy software. Access is restricted to the Gemini API or the Gemini Enterprise Agent Platform, ruling out accidental activation from the consumer Gemini app, leaving the feature strictly in the hands of developers.
Safety features have been beefed up since the previous stand-alone model. Like Gemini 2.5 Computer Use, Gemini 3.5 Flash provides a human-in-the-loop protocol ensuring that explicitly defined “sensitive actions” (like financial transactions) are authorized by a human.
However the update adds two new safeguards:
Note that Google confirms that these most critical safeguards are optional and, as the developer, you are responsible for using them and still bear all the risks if something goes wrong.
Ultimately, the native Google Gemini Computer Use integration offers a significant upgrade over previous models. As part of the main Gemini 3.5 Flash model, there is no premium surcharge to enable computer use. While the newer model is slightly more expensive to use than Gemini 2.5 per million input tokens ($1.50 vs $1.25), it also provides access to a cost-saving context caching feature that can significantly reduce costs over time.
For the millions of developers grinding through repetitive tasks, the benefits will likely offset any small increase in per-token price.
ForbesGoogle Photos Prepares 8 'Adaptive' Filters To Replace Your Editing AppsForbesWhy Snapchat Is Frantically Locking Down Teen AccountsBy Paul Monckton此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。