Overview
- Google made agentic video understanding available to developers through the Gemini API across Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite.
- The company reports the new mode can cut token consumption by up to 88%, lower analysis costs by up to 66%, and improve accuracy by up to 7% on standard video benchmarks.
- Agentic processing replaces fixed-frame sampling with a goal-directed loop that selectively fetches frames, audio clips, or transcript segments so the model inspects only the moments it needs.
- The feature targets hard video tasks such as sub-second moment retrieval, needle-in-a-haystack search across long recordings, anomaly detection, and more accurate counting of actions or objects.
- Developers enable the capability by setting processing to "agentic" in the API with no extra feature fee, and Google plans phased rollouts to the Gemini app and YouTube’s Ask YouTube after positive early-access tests.