A new feature allows Gemini to dynamically explore images, sound and transcription in order to reduce the cost of analyzing long videos.
Announced on September 1, Gemini's agentic video analysis allows the model to search for and then review the passages useful to a question. It replaces the fixed-rate examination with a dynamic selection of sequences and modalities: images, audio or transcription.
Google claims up to 88% fewer tokens consumed and up to 66% cost reduction in its benchmarks. These maximum gains are not guaranteed for every video. The feature is offered via API on Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite, for video files and YouTube links.
Read the full article on the original source
Go to the source →