Computer Use capabilities have been added as a built-in tool to Gemini 3.5 Flash. This allows for the construction of custom agents that can visually perceive, reason, and act across browser, mobile, and desktop environments. While previously offered as a dedicated model for Gemini 2.5, this functionality has now been natively integrated into the main Flash model.
Product LaunchesGoogle DeepMindGemini 3.5 Flash
Computer Use integrated into Gemini 3.5 Flash, enabling cross-platform agent development
This article is a translation. Read the Japanese original
This feature is available via the Gemini API and the Gemini Enterprise Agent Platform. Performance improvements are expected in enterprise automation tasks, such as long-term software testing and knowledge work spanning professional applications.
To mitigate prompt injection risks in agents operating in live environments, adversarial training has been conducted on the target models. Additionally, two types of arbitrary safeguard systems for enterprise use have been released.
Developers are recommended to use these features in combination with secure sandboxing, human-in-the-loop verification, and strict access controls. A best practices document has been published regarding detailed safety measures.
Source: Computer use in Gemini 3.5 Flash (HN 242pt, 167 comments) (HN Search (backfill), 2026-06-25)