Picture artificial intelligence examining images with the diligence of a detective: zooming in, scrutinizing details, and constructing answers backed by visible evidence. With the introduction of Agentic Vision in Gemini 3 Flash, Google is redefining computer vision, transforming passive image processing into an active, investigative process that delivers more accurate and reliable results.
Moving Beyond One-Shot Image Analysis
Traditional AI models often take a single, broad look at an image and attempt to answer questions, sometimes missing important nuances. Agentic Vision introduces a more dynamic approach, known as the “Think, Act, Observe” loop. Instead of making quick guesses, Gemini 3 Flash methodically interacts with images, using code execution to delve deeper into their contents and derive stronger conclusions.
- Think: The model first interprets the query and devises a step-by-step plan for investigation.
- Act: Next, it executes Python code to manipulate images, cropping, rotating, annotating, or analyzing specific features as needed.
- Observe: Each modified image is reintroduced into the model’s context, allowing it to refine its understanding before generating a response.
This iterative strategy consistently increases accuracy, delivering a 5-10% quality boost on vision benchmarks when code execution is enabled.
Real-World Impact: Agentic Vision at Work
Developers and organizations are already seeing tangible benefits from Agentic Vision. Here are a few ways it’s making a difference:
- Zooming and Inspecting: On platforms like PlanCheckSolver.com, Gemini 3 Flash zooms in and analyzes specific image segments, such as fine print on building plans. By programmatically cropping and examining these details, it verifies complex criteria with a 5% improvement in accuracy.
- Image Annotation: Unlike conventional models that simply describe images, Gemini 3 Flash interacts directly with them. For instance, when tasked with counting fingers, it draws bounding boxes and numeric labels, providing a transparent “visual scratchpad” that grounds its answers in the image itself.
- Visual Math and Data Plotting: Agentic Vision excels at interpreting dense tables and performing visual math. Rather than hallucinating results, it uses Python to calculate, plot, and visualize data with professional accuracy, offering insights that can be independently verified.
The Future of Agentic Vision and Gemini
Google is just beginning to unlock the full potential of Agentic Vision. Upcoming enhancements will:
- Automate more code-driven actions, such as image rotations and visual math, making them seamless for users.
- Integrate new tools, including web and reverse image search, to further ground model outputs in real-world context.
- Expand Agentic Vision’s reach to additional Gemini model sizes beyond Flash, broadening its impact.
Developers can start experimenting with Agentic Vision today via the Gemini API in Google AI Studio and Vertex AI, with features rolling out in the Gemini app. Enabling “Code Execution” in AI Studio Playground and consulting the developer documentation makes it easy to explore these advanced capabilities.
A Leap Forward in Visual Reasoning
Agentic Vision represents a major step forward for AI-powered visual reasoning. By empowering models to actively explore, manipulate, and reason about images, Google is paving the way for more trustworthy, contextually aware computer vision applications, benefiting both developers and end-users.
How to get started
Agentic Vision is available today via the Gemini API in Google AI Studio and Vertex AI. It is also starting to roll out in the Gemini app (access by selecting Thinking from the model drop-down). Developers can try the demo in Google AI Studio, or experiment with the feature in the AI Studio Playground by turning on "Code Execution" under Tools. Read the developer docs to learn more for (Vertex AI dev docs).
How Agentic Vision in Gemini 3 Flash is Transforming AI Image Analysis