Details
- Google says Gemini API video analysis is now more accurate and can cost up to 66% less.
- The feature uses agentic video understanding, allowing Gemini to request only the video frames, audio, transcripts, or timeline segments relevant to a prompt.
- Dynamic inspection replaces fixed-rate processing, reducing unnecessary media analysis for long-form video.
- Google reports up to 88% fewer tokens and up to 7% higher accuracy in its benchmark testing.
- The capability is available through the Gemini API, Google AI Studio, and the Gemini Enterprise Agent Platform.
- Supported models include Gemini 4 Argon, Gemini 3.8 Flash, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite, according to Google's developer documentation.
- The feature uses standard Gemini API token pricing, with no separate agentic-video fee.
- Developers can configure video-processing interactions through the GenerateContent API and receive tool-call results as the model examines requested media slices.
Impact
The update could lower the cost of long-form video applications such as search, monitoring, education, and media analysis by shifting from exhaustive frame processing to selective retrieval. It also strengthens Gemini’s position in multimodal developer tooling, although the reported savings and accuracy gains are benchmark-dependent rather than guaranteed for every workload. The main operational trade-off is added model orchestration: applications must handle asynchronous processing requests and returned media segments.