**[Google Adds Video Agent to Gemini: Costs Reduced by Up to 66%]** Beating AI Newsflash: Google has launched Agentic Video Understanding for the Gemini API. This feature allows the model to independently decide which video segments to watch and how to analyze them. In the standard mode, frames are extracted at a fixed rate of 1 FPS and processed in bulk within the context. The new mode autonomously identifies segments along the timeline and selects frames, audio, or transcription based on the question. For fast-moving actions, it can even increase the frame rate for reanalysis. According to Google's internal testing, enabling Gemini 3.7 Flash can reduce token consumption by up to 88%, lower analysis costs by up to 66%, and improve accuracy by up to 7%. It can pinpoint moments shorter than a second or locate specific details in hours-long videos. Technically, this integrates the previously manual "locate first, then analyze" workflow directly into Gemini. Currently, Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite all support this feature. It works with both uploaded videos and YouTube videos, with no additional feature fees. Future updates will integrate this capability into the Gemini App and YouTube's Ask YouTube feature. [Original Link]