Google is adding agent-based video analysis to Gemini 3.7 Flash, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite. Instead of scanning frames at a fixed rate, the system itself selects which segments and resolutions to study.
According to Google, this approach reduces token consumption by up to 88% while simultaneously improving accuracy, particularly when working with multi-hour recordings. These data points were published by The Decoder; the provided package contains only metadata and a synopsis of the material.
The practical implication of this change is the ability to process long videos with a smaller volume of analyzed data. However, the source does not disclose the measurement conditions, test sets, or absolute accuracy values.
editorial commentary
Why it matters
The likely consequence is more cost-effective processing of multi-hour videos; the next verifiable signals will be Google's published testing methodology and results on real materials, while the main source of uncertainty remains the absence of such details in the available synopsis.