Google has unveiled the Gemini 3 Flash, an AI model that enhances visual reasoning capabilities through a new feature known as Agentic Vision. This feature transforms static image processing into an interactive investigation, equipped with the ability to zoom in, inspect, and manipulate images step-by-step. By integrating visual reasoning with code execution, the model can execute Python code for annotating images, parsing complex tables, and visualizing findings, leading to a significant improvement in benchmark performance.
Available through Google AI Studio and Vertex AI, Gemini 3 Flash also provides high-speed performance at a fraction of the cost of its predecessor, the Gemini 3 Pro. This model not only surpasses its forerunners in speed and efficiency but also exhibits improved reasoning and agent capabilities. It has been integrated into Google Anti-gravity and can be accessed via the Gemini CLI, offering downloads for Mac OS, Windows, and Linux in a public preview.
Optimized for low latency and scalability, Gemini 3 Flash’s agentic vision allows developers to create AI agents capable of real-time perception and action. The integration of these capabilities ensures that visual understanding can be incorporated directly into workflows, enhancing efficiency by reducing slow model handoffs. This makes the system particularly valuable for use cases such as interface monitoring, visual task guidance, diagram interpretation, and robotics automation. By addressing the need for tightly coupled perception and action, Google aims to make visual AI more practical and efficient for production systems.

Collections

Google released Gemini 3 Flash, an AI model featuring Agentic Vision that enables interactive image analysis with zoom, inspection, and manipulation capabilities. The model integrates visual reasoning with Python code execution for image annotation, table parsing, and visualization. Available through Google AI Studio and Vertex AI, it offers faster performance and lower costs than Gemini 3 Pro while providing improved reasoning and agent capabilities. The model is optimized for low-latency applications like interface monitoring, visual task guidance, diagram interpretation, and robotics automation.

Google Launches Advanced Visual Reasoning with Gemini 3 Flash