What Happened
Google announced Agentic Vision as a capability in Gemini 3 Flash that treats vision understanding as an active, tool-augmented investigation.
Why It Matters
This is an explicit “vision + tools” milestone: instead of one-shot image understanding, the model can plan, zoom/inspect, and ground answers via code execution—an important direction for document analysis, forensics, and agentic UX.
Technical Details
- Approach: Visual reasoning paired with code execution to ground outputs in evidence
- Availability: Announced for Gemini API surfaces (Google AI Studio / Vertex AI) and rollout paths