Google Launches Agentic Video Understanding for Gemini Flash Models, Cutting Video Tokens by Up to 88%

Video has been the costliest modality to motive over. A Gemini mannequin handed a 90-minute lecture has, till now, ingested the entire thing at a set one body per second, whether or not the query was ‘summarize this’ or ‘what time does the speaker change to the pricing slide?’ That single-pass design forces a nasty commerce: pay for the total timeline in context, or pre-chunk the video and threat dropping the element that mattered.

This week, Google launched agentic video understanding throughout its Flash fashions. Instead of ingesting the timeline, Gemini navigates it deciding what to watch, at what body fee, and thru which modality. Google experiences up to 88% fewer tokens, up to 66% decrease value, and up to 7% increased accuracy on customary video benchmarks.

Is it deployable? Yes, however solely as a hosted API function. There are not any open weights and nothing to self-host. It ships by the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform, works with each file uploads and public YouTube URLs, and payments at customary Gemini API token pricing with no extra function payment.

What truly modified

Static processing, nonetheless the default on each Gemini mannequin, extracts frames at 1 FPS in a single go, processes audio at 1 Kbps single channel, and inserts timestamps each second. Agentic processing replaces that with a loop. The mannequin pairs its personal reasoning with native video instruments to search, scan, and examine goal segments throughout frames, audio, and transcripts, loading solely what the immediate requires. Developers may already assemble this by hand; the change is that Gemini runs the loop internally, which is the place the event overhead disappears.

Across Google’s evaluations, Gemini 3.7 Flash with agentic understanding lands on the accuracy-to-cost Pareto frontier for video evaluation among the many fashions examined. The effectivity beneficial properties consider long-form content material, from 10-minute how-to guides to multi-hour recordings.