Google Launches Agentic Video Analysis for Gemini Flash

Google has rolled out a new agentic video understanding capability for its Gemini 3.7 Flash, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite models. Available immediately, this feature allows the models to dynamically scan, search, and inspect specific video segments rather than processing files at a fixed frame rate. According to Google, this shift from static analysis to an active agentic loop improves accuracy by up to 7 percent while cutting token consumption by up to 88 percent and reducing overall costs by up to 66 percent.

Traditionally, video analysis relies on static ingestion at a default rate of 1 frame-per-second, which developers can adjust via the API. This approach often forces a compromise between high token costs and losing critical details in long-form footage. The new agentic system allows Gemini to actively decide what parts of a video to watch, at what speed, and through which modality, whether analyzing visual frames, audio, or transcripts. On the LongVideoBench benchmark, this method places Gemini 3.7 Flash at the accuracy-to-cost pareto frontier.

Developers can leverage this feature for complex tasks such as sub-second moment retrieval, long-form needle-in-a-haystack search, anomaly detection, and precise action or object counting. The capability is accessible today for video uploads and YouTube videos via the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform. To activate it, developers simply need to set their processing configuration to "agentic" in the API. The feature is billed at standard Gemini API token pricing with no additional fees.

Google plans to expand the reach of this technology beyond developer platforms. The agentic video feature will soon roll out to general users within the Gemini app for Flash and Flash-Lite models. Furthermore, Google intends to integrate the capability into YouTube's "Ask YouTube" watch-page feature in the coming months to provide higher-quality, visually grounded answers.