Gemini 3.5 Flash Gets Native Computer Use — AI Agents Now Navigate Any Screen Autonomously

On June 24-25, 2026, Google announced that computer use is now a built-in native tool inside Gemini 3.5 Flash — the same production model already used by enterprise developers for function calling, Search grounding, and Maps integration.

Key details:

• Computer use was previously accessible only through a standalone Gemini 2.5 Computer Use model; making it native to Gemini 3.5 Flash eliminates routing complexity

• Performance: Gemini 3.5 Flash with computer use scores 78.4 on OSWorld-Verified benchmark — matching GPT-5.5's score of 78.7 (0.3 point difference) at roughly one-third the per-token cost

• A Gemini 3.5 Flash agent can receive a high-level task ('Book the cheapest flight from Newark to Toronto on July 18 and add it to my calendar'), open a browser, navigate booking sites, compare prices, fill passenger information, complete the booking, and create the calendar event — all without human intervention

• Enterprise deployments can scope the agent's screen access to specific application windows or browser tabs, providing sandboxed execution

• Integration with Search and Maps means a computer use agent that can also search the web and query mapping data — combining world knowledge, location intelligence, and interface navigation in a single model call chain

Available through the Gemini API and Gemini Enterprise Agent Platform.

Why It Matters: This is the most practically significant AI agent capability advancement for enterprise automation since Claude Code's 2025 release. Enterprise automation can now reach workflows that API-based automation cannot — specifically legacy applications with no API, complex web interfaces, and desktop software never programmed for machine interaction.