Introducing agentic video understanding with Gemini
deepmind ·
Google DeepMind has launched agentic video understanding for its Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite models, allowing the models to dynamically scan video segments instead of using static frame-rate processing. This feature reduces token consumption by up to 88%, lowers costs by up to 66%, and improves accuracy by up to 7% on standard benchmarks. It is available through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform, and can be enabled by setting the API processing mode to 'agentic'. The technology enables sub-second moment retrieval, long-form needle-in-a-haystack search, anomaly detection, and accurate counting of actions and objects. Early access partners reported strong performance, and Google plans to roll the feature out to the Gemini app and later to YouTube's 'Ask YouTube' feature.
deepmind