Agentic Video Understanding in Gemini

Agentic video understanding introduces a dynamic processing mode for video content, allowing systems to autonomously determine optimal viewing parameters. Key features include: • Dynamic frame rate adjustment • Modality selection for viewing • Significant reduction in token usage (up to 88%) • Substantial cost savings (up to 66%) • Enhanced accuracy, especially for extended videos (up to 7% boost) This innovative capability empowers models to efficiently analyze video by making contextual decisions about what to observe and at what pace, moving beyond fixed frame rate limitations. It dramatically optimizes resource consumption, leading to lower operational costs while improving the precision of video comprehension. The approach is particularly beneficial for processing lengthy video content, where traditional methods can be inefficient. Currently accessible via the Gemini API in products like AI Studio and the Gemini Enterprise Agent Platform, users can activate this feature by setting the processing mode to "agentic." This advanced video processing is available at standard pricing, ensuring accessibility without additional fees. It's ideal for developers and enterprises focused on large-scale video analysis, content moderation, or any application requiring deep understanding of visual data with optimized performance.