Introducing agentic video understanding with Gemini

Google has launched agentic video understanding across its latest Gemini models—Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite—delivering significant improvements in accuracy, token efficiency, and cost for video analysis. Unlike traditional static video processing, which ingests footage at a fixed frames-per-second rate, agentic video understanding enables Gemini to take an active, goal-directed role in analyzing video. The model dynamically decides what to watch, at what speed, and through which modality (frames, audio, or transcript), using an internal tool to fetch only the most relevant moments and signals.

According to Google, the new approach reduces token consumption by up to 88%, cuts analysis costs by up to 66%, and improves accuracy by up to 7% across standard video analysis benchmarks. These gains are especially notable for long-form content such as how-to guides, lectures, and multi-hour recordings, where static processing traditionally forced developers to choose between high costs and lost details. Gemini 3.7 Flash with agentic understanding reportedly offers the best overall quality and the best accuracy-to-cost balance among tested models.

The feature unlocks several new capabilities, including sub-second moment retrieval for precise editing, long-form needle-in-a-haystack search, anomaly detection through dynamic resampling, and accurate counting of actions and objects over time. Early access partners reported strong performance, and the feature is now available via the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform for both video uploads and YouTube videos. Developers can enable it by setting the processing parameter to "agentic." Google also plans to roll out the capability to Gemini app users across Flash and Flash-Lite models, and to power YouTube's "Ask YouTube" feature in the coming months.

(Source：Gemini)