Google has announced that "computer use," the ability to view a screen and operate it, is now a built-in tool in its Gemini 3.5 Flash model[1]. The capability was previously offered as a standalone model, but it is now integrated into the flagship Flash model, letting developers build agents that act autonomously across browser, mobile and desktop environments[1]. It is available through the Gemini API and the Gemini Enterprise Agent Platform[1].
Computer Use Becomes a Built-In Tool in the Flagship Model
Computer use is now a tool that ships with Gemini 3.5 Flash by default[1]. Google describes it as the company's "best performance yet" for agentic computer-use tasks[1].
Gemini has long been strong at function calling and at using built-in tools such as Search and Maps grounding[1]. With the added ability to operate a screen directly, developers can now use 3.5 Flash to build agents that see, reason and take action across browser, mobile and desktop environments[1]. Google says this improves performance on long-horizon work and enterprise automation[1].
What It Can Do
As expected use cases, Google points to multi-step automation such as continuous software testing and knowledge work across professional applications[1]. In the company's own examples, 3.5 Flash analyzes the Gemini app's screen and returns a categorized list of features, and it audits its own documentation to check for accessibility issues[1].
Computer use here works as a loop: the model is given the state of the screen, it returns an action such as "click," "type" or "scroll," and the result is fed back to it as a fresh view of the screen — repeating until the task is done[2][3]. It helps to think of the AI carrying out the same steps a person would with a mouse and keyboard.
From a Standalone Model to a Built-In Capability
The feature was originally offered as a dedicated model, "Gemini 2.5 Computer Use," released in October 2025[2][3]. That preview model was built on Gemini 2.5 Pro and was designed mainly for operating web browsers and mobile apps; it was not yet optimized for controlling a desktop OS[3].
Now Google has taken that role out of a separate model and built it into the flagship 3.5 Flash[1]. The environments Google lists include the desktop, broadening the range it can handle[1]. To try it, Google points to a demo environment hosted by the browser-automation company Browserbase, along with a reference implementation and documentation available through the Gemini API[1].
Guarding Against Prompt Injection
An agent that operates a screen automatically is convenient, but it carries the risk of "prompt injection," in which malicious instructions are slipped in from the outside. Google says it applied targeted adversarial training to computer use in 3.5 Flash to address such attacks[1].
It also offers two optional safeguard systems for enterprises[1]. One requires explicit user confirmation before sensitive or irreversible actions[1]. The other automatically stops a task when an indirect prompt injection is detected[1]. Google frames these as part of a "defense-in-depth" approach and encourages developers to combine them with secure sandboxing, human-in-the-loop verification and strict access controls[1].
Summary
Google has folded computer use, which began as a standalone model, into its flagship 3.5 Flash and positioned it as a built-in feature for building agents that span browser, mobile and desktop[1]. Alongside the performance gains, the emphasis on prompt-injection defenses and enterprise safeguards signals an effort to move these agents closer to real, day-to-day business use[1].
Source: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-computer-use-model/
Source: https://ai.google.dev/gemini-api/docs/models/gemini-2.5-computer-use-preview-10-2025
