According to CoinMeta, Google has launched agentic, video, and understanding features on Gemini API, allowing models to independently select video segments for inspection. The new mode can search for segments along the timeline and choose images, audio, or transcribed text based on specific questions. Google's internal tests show that after enabling these features (version 3.7 of Gemini and flash), token consumption decreased by up to 88%, analysis costs dropped by up to 66%, and accuracy improved by about 7%. This feature supports uploading both regular videos and YouTube videos, without any additional fees.