Google Gemini's new agent-based video analysis cuts token usage by up to 88 percent

Google is adding agent-based video analysis to Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. Instead of scanning videos frame by frame at a fixed rate, the model decides on its own which segments to examine and at what resolution. Google says this cuts token usage by up to 88 percent while improving accuracy, especially for multi-hour footage. The article

The top 3

  1. Tech Giants' AI Supremacy Race: This section examines the leading positions of major tech companies in both overall brand value and their significant investments and influence in the rapidly evolving AI landscape.
  2. Leading AI Models & Efficiency Breakthroughs: This section highlights the most advanced AI models and the innovative techniques used to enhance their efficiency, a crucial aspect given the high computational costs of large models.
  3. Advancements in AI Video Analysis: This section explores the historical journey and modern applications of AI in understanding and processing video content, highlighting key breakthroughs and practical uses.

Sources

Open the full topic