Overview

• Gemini 2.5 Computer Use is now in public preview through the Gemini API on Google AI Studio and Vertex AI, with demos hosted on Browserbase.
• Built on Gemini 2.5 Pro, the model drives UI actions like clicking, typing, scrolling and drag‑and‑drop by iterating on screenshots and recent action history.
• The model is optimized for web browsers, shows strong promise on Android control benchmarks, and is not yet optimized for desktop OS‑level control.
• Google reports the model outperforms leading alternatives on multiple web and mobile control benchmarks while delivering lower latency.
• Safety features include model‑level safeguards, a per‑step safety service that reviews each action, and system instructions to require refusals or user confirmations for high‑risk steps, with versions already used for UI testing and powering Project Mariner, the Firebase Testing Agent and agentic features in AI Mode.