Google Launches Agentic Video Understanding in Gemini Slashing Token Costs by 66%
Google introduced agentic video understanding across Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite via API. By dynamically scanning video segments rather than ingesting fixed frames, it cuts tokens by up to 88% and costs by 66%.
Impact: High
Why it matters
You can slash video processing API costs by setting processing='agentic' in Google AI Studio without changing your underlying prompts.
TL;DR
- 01Agentic video understanding reduces token consumption by up to 88% and cost by up to 66%.
- 02Gemini dynamically searches, scans, and resamples frames instead of ingesting fixed 1 FPS streams.
- 03Available immediately in Google AI Studio across Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite.
Key facts
- Token reduction (self-reported)
- Up to 88%
- Cost reduction (self-reported)
- Up to 66%
- Accuracy increase (self-reported)
- Up to 7%
- Supported models
- Gemini 3.7 Flash, 3.6 Flash, 3.5 Flash-Lite
Dynamic Sampling Replaces Static Ingestion
Traditional video processing in multimodal models operates statically at a fixed rate (such as 1 frame per second). Agentic video understanding gives Gemini an internal loop with native video tools to dynamically inspect visual frames, audio channels, and transcripts. The model chooses what sections to inspect, at what frame rate, and through which modality, fetching only required signals.
Benchmark Numbers and Token Savings
According to Google's self-reported data on LongVideoBench, activating agentic video understanding yields:
- Up to 88% reduction in token consumption
- Up to 66% reduction in analysis costs
- Up to 7% boost in retrieval and reasoning accuracy
Gemini 3.7 Flash with agentic mode establishes an efficiency Pareto frontier, outperforming static 1 FPS processing even on multi-hour recordings and split-second state transitions.
Integration and API Configuration
The capability is available immediately for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite via the Gemini API in Google AI Studio and Gemini Enterprise Agent Platform. It applies to uploaded video files and YouTube URLs. Developers can enable the feature by setting the processing parameter to "agentic" in their API interaction configuration, billed at standard model token rates without separate feature fees.
Try it in 2 minutes
interaction = client.interactions.create(
model="gemini-3.7-flash",
processing="agentic",
contents=[video_file, "What are the key announcements?"]
)python
✓ When to use
- When analyzing long-form video, tutorials, lectures, or surveillance feeds where static FPS wastes millions of tokens.
- When needing sub-second precision to pinpoint frame changes or count fast actions.
✕ When NOT to use
- When analyzing short 2-3 second video clips where static 1 FPS ingestion is already minimal in token cost.
- For purely audio-only processing where native audio transcription models are cheaper.
What to do today
- Add processing='agentic' to your Gemini API client interactions.
- Benchmark token savings on your long-form video workloads using Gemini 3.7 Flash.
Sources