Google Gemini slashes video analysis token usage by 88%

google's gemini models, including gemini 3.7 flash, 3.6 flash, and 3.5 flash-lite, are being equipped with agent-based video analysis. this feature allows the models to dynamically search through video footage instead of scanning it frame by frame in a fixed pattern. according to google, this approach saves up to 88 percent in tokens and cuts costs by 66 percent, while improving accuracy. the feature is now available through the gemini api for finding specific scenes in long videos, with integration into the gemini app and youtube planned for a later date.