Google launches agentic video understanding in Gemini Flash models

Google on 1 September 2026 launched agentic video understanding across Gemini 3.7 Flash, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite. The feature is available now for video uploads and YouTube videos through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform.

The post, by Google DeepMind’s Rohan Doshi and Mario Lučić, compares the new path to “agentic vision,” which already combined code execution with Gemini’s native image understanding. For video, Gemini uses native video tools in an agentic loop instead of ingesting a stream at a fixed frame rate.

What Google claims on quality and cost

Static processing, Google wrote, ingests video at a fixed frames-per-second rate, defaulting to 1 FPS and adjustable via the API. Agentic video understanding “pairs the model’s core reasoning with native video tools to dynamically search, scan, and inspect target video segments across visual frames, audio, and transcripts.”

On “standard video analysis benchmarks,” Google said Gemini models with agentic video understanding reduce analysis costs by up to 66% and token consumption by up to 88%, while improving accuracy by up to 7%. The company said those efficiency gains are strongest on long-form video, from 10-minute how-to guides to 90-minute lectures and multi-hour recordings, where static processing forces a choice between high token cost and dropping detail.

Google said Gemini 3.7 Flash with agentic understanding offers the best overall quality among the three supported models and the best quality-to-cost combination, placing it at the “accuracy-to-cost pareto frontier among tested models for video understanding.”

How developers turn it on

Instead of watching every frame at a constant rate, Gemini decides what to watch, at what speed, and through which modality (frames, audio, or transcript), “fetching only the moments and signals needed.” Google said developers could previously do this by hand; the new path lets the model invoke an internal tool to load the relevant part of the file.

The company listed four example capabilities: sub-second moment retrieval for tight cuts; long-form needle-in-a-haystack search across multi-hour video without consuming millions of tokens; anomaly detection by resampling interesting windows at higher FPS; and counting repeated actions and distinct objects over time.

There is no extra feature fee. Agentic video understanding uses standard Gemini API token pricing. To enable it, Google said developers set processing to "agentic" in the API configuration. A sample in the post uses model="gemini-3.7-flash" with a YouTube URI and "processing": "agentic".

Where it shows up next

The same blog post said the feature will roll out to all users in the Gemini app across Flash and Flash-Lite models “soon.” In the coming months, Google said agentic video understanding will also power YouTube’s “Ask YouTube” feature on the watch page, “leveraging Gemini to deliver higher-quality answers grounded in the visuals.”

This is not a new Gemini numbered model. It is a processing mode on 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite, live in AI Studio and the Gemini Enterprise Agent Platform as of 1 September 2026. Unofficial reports of a Gemini 3.8 Flash launch are not confirmed in this post and are not part of Google’s announcement.

Source: Google, 1 September 2026.