Computer use in Gemini 3.5 Flash (blog.google)

Gemini 3.5 Flash has introduced built-in computer use capabilities, enhancing its performance for agentic computer tasks. This advancement marks a significant upgrade from the standalone Gemini 2.5 model, allowing developers to create more effective custom agents that can interact seamlessly across various environments—desktop, mobile, and browser. The integration supports complex tasks such as long-horizon automation and enterprise applications, enabling improved functionalities like continuous software testing and better knowledge management. To enhance safety during operations, Gemini 3.5 Flash implements targeted adversarial training to address prompt injection risks. It includes two optional enterprise safeguards to ensure user confirmation on critical actions and stops tasks if such risks are detected. Developers are encouraged to incorporate these features alongside secure practices like sandboxing and human oversight. These updates not only bolster the robustness of AI interactions in real-time environments but also increase the overall trustworthiness of AI systems in enterprise applications.