On June 24, 2026, Google DeepMind announced that Gemini 3.5 Flash now has a built-in computer use tool that directly operates browser, mobile, and desktop environments. The screen-control capability that was previously available only as a standalone Gemini 2.5 model is now integrated into the Flash model itself.

What Changed

On June 24, 2026, Google DeepMind said it built computer use into Gemini 3.5 Flash as a native tool. Computer use is a capability that lets an agent see, reason over, and then take actions—such as clicking and typing—across browser, mobile, and desktop environments. It was previously available only as a separate Gemini 2.5 computer use model, but it is now integrated into the Flash model itself.

Where You Can Use It

Google said it offers Gemini 3.5 Flash computer use through the Gemini API and the Gemini Enterprise Agent Platform. As early adoption and testing references, Browserbase, Browser Use, and UiPath were cited in areas like continuous software testing and automation across professional applications. For performance, an OSWorld benchmark comparison was presented alongside the launch.

What the Safeguards Are

Google said it applied targeted adversarial training to Gemini 3.5 Flash computer use to reduce prompt-injection risk. It added an optional enterprise safeguard that requires explicit user confirmation for sensitive actions, and it automatically stops a task when indirect prompt injection is detected. Because the agent operates real systems, Google framed security controls as a core part of the feature.

Why It Matters

Gemini 3.5 Flash is Google's low-latency, low-cost line, so building computer use into the base model matters because it lowers the unit cost of agent automation. Handling screen control within a single Flash model, without routing through a separate dedicated model, lowers the barrier to long-horizon tasks and enterprise workflow automation. As the computer-use agent race intensified in 2026, Google is differentiating on the axes of speed and cost.

Summary

On June 24, 2026, Google DeepMind built a computer use tool that operates browser, mobile, and desktop into Gemini 3.5 Flash, offering it through the Gemini API and the Enterprise Agent Platform. It framed safeguards—such as prompt-injection defenses—as a core part of the feature.

Source: Google DeepMind