Gemini adds targeted inspection for long-video analysis

Google launched agentic video understanding on 1 September for Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite. Instead of treating a video only as frames sampled at a fixed rate, the feature lets a model search, scan and inspect selected segments using visual, audio and transcript tools.

The capability supports uploaded video and YouTube material through the Gemini API in AI Studio and the enterprise platform. Google reported token reductions up to 88% and cost reductions up to 66% on its stated evaluations. Those maxima depend on the task and recording. The approach is particularly relevant when a question concerns a brief event inside a long lecture or recording, where repeatedly processing every frame can spend tokens on unrelated material.