谷歌Gemini代理式视频分析大幅降低令牌消耗

AI 摘要

谷歌为其Gemini Flash模型新增了基于代理的视频分析功能，该功能不再按固定帧率扫描视频，而是由模型自主决定分析哪些片段及方式。据谷歌称，这可将令牌使用量减少高达88%，成本降低66%，并提高准确性。该功能现已在Gemini API中提供，用于在长视频中查找特定场景，未来将集成到Gemini应用和YouTube中。

正文节选

Google Gemini's new agent-based video analysis cuts token usage by up to 88 percent Key Points - Google is equipping its Gemini Flash models with agent-based video analysis that dynamically searches through video footage instead of scanning it frame by frame in a fixed pattern. - The models autonomously decide which sections of a video to analyze and how, which, according to Google, saves up to 88 percent in tokens, cuts costs by 66 percent, and improves accuracy. - The feature is now available

发布时间：2026-09-02 16:21

抓取时间：2026-09-02 17:18

来源机构：THE DECODER