Documented agentic video understanding on /gemini-api/docs/video-understanding and /gemini-api/docs/generate-content/video-understanding: for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash Lite, the model can dynamically navigate the timelin…
Changelog
Video understanding
Documented agentic video understanding on /gemini-api/docs/video-understanding and /gemini-api/docs/generate-content/video-understanding: for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash Lite, the model can dynamically navigate the timeline and load only needed transcript, frames, and audio (vs default static 1 FPS processing), with guidance on when to choose each mode and how to set processing / media_processing.
Tokens
Clarified video token rates on /gemini-api/docs/tokens and /gemini-api/docs/generate-content/tokens: the fixed per-second rate applies to static processing; added a Video token usage by processing mode section covering static vs agentic calculation, typical savings on long-form content, and where agentic tokens appear in usage metadata.
On /gemini-api/docs/optimization, added guidance to reduce long-form video input costs with agentic processing (up to 88% fewer input tokens), including API parameter names and when static may still be preferable for short, latency-sensitive clips.
On /gemini-api/docs/pricing, noted that agentic video understanding bills from variable tokens actually loaded rather than full video length.