Google Launches Agentic Video Understanding for Gemini Flash Models, Cutting Video Tokens by Up to 88%

Google has introduced an agentic video understanding feature for Gemini Flash models that allows the AI to navigate video content by selecting specific segments relevant to a prompt. Instead of processing every frame at a fixed rate, the model now targets only the necessary portions of the footage. This approach reduces video token consumption by as much as 88 percent, lowering the computational requirements for analyzing long or complex video files.