Google DeepMind has launched a new agentic video understanding feature for its Gemini models, including Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. This capability enables the models to dynamically scan video segments, improving accuracy while significantly reducing token consumption and costs.
According to the company, the feature cuts token usage by up to 88% and reduces costs by up to 66%, while boosting accuracy by up to 7% on standard video analysis benchmarks. The efficiency gains are especially notable for long-form videos, such as 10-minute tutorials, 90-minute lectures, and multi-hour recordings.
Unlike traditional static processing, which ingests video at a fixed frame rate (default 1 FPS), agentic video understanding pairs Gemini's core reasoning with native video tools to search, scan, and inspect target segments across visual frames, audio, and transcripts. This allows the model to fetch only the moments and signals needed, reducing development overhead.
The feature is available today for video uploads and YouTube videos via the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform. Developers can enable it by setting processing to "agentic" in the API configuration, with no additional feature fee.
Google also plans to roll out agentic video understanding to the Gemini app and YouTube's 'Ask YouTube' feature in the coming months, bringing the efficiency and quality improvements to billions of users.