Gemini 3.5 Flash Computer Use Integration

Google DeepMind has integrated computer use as a native tool within Gemini 3.5 Flash, allowing developers to build agents capable of interacting across browser, mobile, and desktop platforms. This integration moves computer use from a standalone model (previously Gemini 2.5) into the main Gemini Flash model to improve performance for long-horizon and enterprise automation tasks.

Native Computer Use Capabilities

Gemini 3.5 Flash now natively supports computer use, enabling the creation of custom agents that can perceive, reason, and execute actions across various digital environments. This capability complements existing Gemini strengths in function calling and grounding via tools like Search and Maps.

Key applications for this native integration include:

• Enterprise Automation: Streamlining knowledge work across professional applications.
• Software Testing: Executing continuous software testing workflows.
• Cross-Platform Interaction: Operating seamlessly across desktop, mobile, and browser interfaces.

Safety and Security Framework

To address prompt injection risks associated with agents in live environments, Google DeepMind employs targeted adversarial training for Gemini 3.5 Flash. The lab advocates for a "defense-in-depth" strategy, recommending that developers combine native features with secure sandboxing, strict access controls, and human-in-the-loop verification.

For enterprise users, two optional safeguard systems are available:

• Explicit User Confirmation: Requires a human to approve sensitive or irreversible actions.
• Indirect Prompt Injection Detection: Automatically terminates tasks if an indirect prompt injection is identified.

Availability and Implementation

Computer use in Gemini 3.5 Flash is available through the Gemini API and the Gemini Enterprise Agent Platform. Developers can access a reference implementation via GitHub or test the capabilities in a demo environment hosted by Browserbase.