Google launched agentic video understanding for Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite. The model now decides which parts of a video to watch, at what speed, and whether to use frames, audio or transcript. Previously it read every video at a fixed frame rate. Google reports lower cost and token use with better accuracy on standard video benchmarks. Gains are largest on long videos such as lectures and multi-hour recordings. New uses include pinpointing split-second moments, counting actions and objects, and spotting anomalies. It works today on uploads and YouTube videos through the Gemini API and the Gemini Enterprise Agent Platform. Standard token pricing applies with no extra fee. Google says the feature will reach the Gemini app soon and YouTube's Ask feature in the coming months.
What changed
Video was ingested at a fixed frame rate, default 1 frame per second.
What it unlocks
Letting the model search a multi-hour video and load only the needed moments.
- up to 88% fewer tokens
- up to 66% lower cost
- up to 7% better accuracy
What you need to act on it
- Gemini API access via Google AI Studio or Gemini Enterprise Agent Platform
- set processing to "agentic" in the API configuration
Sources