Gemini adds agentic video understanding with lower analysis cost

Google launched agentic video understanding on September 1 for Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite. Instead of processing every frame in the same way, the model can inspect the moments relevant to a task.

Google reports up to 88% lower token use, up to 66% lower cost and quality gains of up to 7% in its evaluations. Example tasks include moment retrieval, anomaly detection and counting.

For a searchable case library, this points toward automatic scene and action indexing with answers tied to exact moments. Results will vary by footage and task.