具身智能观察

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

产业动态

来源:THE DECODER发布时间待核实

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 Google Gemini's new agent-based video analysis cuts token usage by up to 88 percent appeared first on The Decoder.

Google Gemini's new agent-based video analysis cuts token usage by up to 88 percent | 具身智能观察