Rather than decoding a video timeline uniformly at a fixed frame rate, the model can inspect transcripts, retrieve targeted frame sequences at adaptive rates, and extract audio as needed through an iterative server-side tool loop. The guide lists support for Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite, enabled with processing="agentic". Google reports up to 88% fewer tokens and up to 66% lower analysis cost for long-form media compared with static 1 FPS processing, and up to 7% higher accuracy on fine-grained visual reasoning tasks. These are vendor-reported benchmark maxima, not guaranteed outcomes for every workload. Source: https://aistudio.google.com/learn/agentic-video-understanding-with-gemini
Google launched agentic video understanding for Gemini Flash models, cutting long-form video costs up to 66%.
In one minute
Google’s Gemini API developer guide documents an agentic video-processing mode for Gemini Flash models.
What changed: The developer guide quantifies the workflow’s gains: up to 88% fewer tokens, 66% lower long-form analysis costs, and 7% higher accuracy.