Google has launched agentic video understanding across its Gemini Flash models, replacing the previous static processing method with a system that navigates video timelines on demand. This change allows the model to decide what to watch, at what frame rate, and through which modality, rather than ingesting the entire timeline at a fixed one frame per second.
- The new approach reduces token usage by up to 88% and lowers costs by up to 66% while improving accuracy on standard video benchmarks by up to 7%.
- Gemini 3.7 Flash with agentic understanding now sits on the accuracy-to-cost Pareto frontier for video analysis among tested models.
- The feature is enabled via a single "processing" field in the API and supports both file uploads and public YouTube URLs.
- Token accounting distinguishes between navigation reasoning (thought tokens) and loaded media (tool-use tokens).
This update allows developers to handle long-form content more efficiently without the development overhead of manually assembling processing loops, though static processing remains recommended for clips under five minutes.