Google DeepMind Launches Agentic Video Understanding with Gemini 3.7 Flash

On September 1, 2026, Google DeepMind launched the agentic video understanding feature within the Gemini 3.7 Flash model. This feature achieves a token consumption reduction of up to 88% and a cost reduction of up to 66%, while improving accuracy by up to 7%, according to Senior Product Manager Rohan Doshi. The model can now analyze video lengths ranging from 10 minutes to multi-hour recordings at a default frame rate of 1 FPS. Doshi noted that activating agentic video understanding positions Gemini 3.7 Flash at the accuracy-to-cost pareto frontier for video analysis. This follows the recent enhancements in Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, which focused on video analysis capabilities. Practitioners can leverage these improvements for more efficient video processing in their applications, as detailed on the Google DeepMind Blog.