Gemini 3.5 Flash can now browse and operate computers
What's the story
Google has launched its latest AI model, the Gemini 3.5 Flash, with a unique feature called "computer use." This new capability allows developers to create agents that can interact across different platforms like mobile and desktop. The feature was previously only available as a standalone Gemini 2.5 computer use model but is now natively integrated into the main Gemini Flash model for improved performance on agentic computer use tasks.
Enhanced capabilities
Major upgrade for agentic computer use tasks
The Gemini 3.5 Flash model is a major upgrade over its predecessor, specifically for agentic computer use tasks. It builds on the already impressive capabilities of the Gemini model in function calling and using built-in tools like Search and Maps grounding. With this new integrated capability, developers can now create custom agents that can see, reason, and act across different environments such as browsers, mobile devices, and desktops.
Enterprise applications
Enterprise automation and long-horizon tasks
The new integrated capability of Gemini 3.5 Flash also promises improved performance for long-horizon and enterprise automation tasks, such as continuous software testing and knowledge work across professional applications. Developers and enterprises can start using this feature through the Gemini API and the Gemini Enterprise Agent Platform. The model also uses computer use to analyze the Gemini app and return a categorized list of features, making it even more useful for businesses.
Safety measures
Addressing prompt injection risks
To address prompt injection risks for agents operating in live environments, Google DeepMind has introduced targeted adversarial training for computer use in Gemini 3.5 Flash. The company is also releasing two optional enterprise safeguard systems that allow enterprises to require explicit user confirmation for sensitive or irreversible actions and automatically stop tasks if an indirect prompt injection is detected.