Google Cuts Video Token Costs with Agentic Video Understanding

Google has made significant strides in video analysis by launching agentic video understanding for its Gemini Flash models. This innovation allows Gemini to navigate a video timeline, deciding what to watch and at what frame rate, rather than ingesting the entire timeline in one pass.

The traditional method of processing videos resulted in a bad trade-off: either pay for the full timeline in context or pre-chunk the video, risking dropped details. Agentic processing replaces this single-pass design with a loop that pairs Gemini's reasoning with native video tools to search and inspect target segments across frames, audio, and transcripts.

According to Google, agentic processing leads to up to 88% fewer tokens, up to 66% lower cost, and up to 7% higher accuracy on standard video benchmarks. The new feature is deployable only as a hosted API feature, with no open weights or self-hosting capabilities.