What Google Actually Shipped on September 1
Google added agentic video understanding to three models in its Gemini Flash lineup on September 1, 2026: Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite, according to Google's own announcement and independent reporting from Android Authority. The feature is a processing technique, not a new model or a new product line, and it should not be confused with Gemini 3.8 Flash, a separate model that Google released a day later, on September 2, with its own pricing and release-cadence questions already covered elsewhere. Agentic video understanding changes how these three Flash models read video, while leaving the underlying model weights and their existing capabilities in place.
How the Model Decides What to Watch
The mechanism replaces fixed-frame-rate scanning with a model that decides, segment by segment, what to inspect, how fast to move through it, and which modality to use. Older video-understanding approaches sample a video at a set number of frames per second regardless of content, so a static two-hour security recording gets the same frame-by-frame treatment as two minutes of fast action. Google's agentic approach instead searches a video for the segments that matter and inspects only those, switching between frames, audio, and transcripts depending on what the question requires. That targeted search is what unlocks sub-second moment retrieval in long-form and multi-hour video, detection of visual anomalies or artifacts, and counting of objects or actions across a clip, tasks that uniform frame sampling handles far less efficiently.
The Numbers, Compared
Google publishes three headline figures for the new processing mode, and none of them describes a new price tier.
| Metric | Google's claim |
|---|---|
| Token consumption | Up to 88% lower |
| Processing cost | Up to 66% lower |
| Quality/accuracy | Up to 7% higher |
All three figures come from Google's own testing and apply at standard Gemini API token pricing; there is no additional fee for turning agentic processing on. Developers enable it by setting the API's processing parameter to agentic, available now in Google AI Studio and the Gemini Enterprise Agent Platform.
Why This Is a Workload-Economics Story, Not a Pricing Story
For EU and UK teams running video-heavy AI workloads, an 88% cut in token consumption changes whether large-scale video analysis is affordable at production scale, not merely how much a single query costs. This is a distinct story from Gemini 3.8 Flash's separate pricing and cadence changes: Google is bundling an efficiency technique into the existing Flash line at no extra fee, rather than selling a new tier. Before September 1, the actual blocker on many video-analysis projects, continuous security-camera review, multi-hour footage triage for content moderation, media monitoring across broadcast archives, ad-creative quality checks across a video library, was often the token cost of frame-by-frame ingestion rather than any limit on model capability. An 88% reduction in that ingestion cost moves several of those use cases from a line item that gets cut in a budget review to one that a team can actually run continuously.
Servola Journal
We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.
Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.
If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.
Read next: Google's Watermark Opt-Out Tests the EU AI Act | Google Assistant Shuts Down September 4 for Gemini
