Google has announced the release of agentic video understanding across its Gemini Flash models, moving away from single-pass, fixed-frame-rate processing. Available via the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform, the feature uses an internal reasoning loop to navigate video content, dynamically selecting specific frames, audio segments, and transcripts based on user prompts.
According to Google, this targeted inspection reduces token consumption by up to 88% and lowers processing costs by up to 66%, while delivering up to 7% higher accuracy on standard video benchmarks. The efficiency improvements are particularly pronounced on long-form content, such as multi-hour recordings and lengthy instructional videos.
The feature is currently available only as a hosted API service with no open-weights or self-hosting option. Developers can integrate it using standard Gemini API token pricing, mixing agentic and static modes within single multi-video requests.
Why it matters
Drastically lowers API costs for startups building video processing, transcription, or search applications.
Eliminates the overhead of building custom chunking and retrieval loops for long-form video analysis.
Locks developers into Google’s hosted API ecosystem due to the lack of open-weights support.
Source: marktechpost.com
