Google Launches Agentic Video Understanding for Gemini, Slashing Analysis Costs by 66%

Google has launched agentic video understanding across its latest Gemini models, fundamentally changing how AI analyzes video content.1

1

2

2

This new capability is now available for Gemini 3.7 Flash, Gemini 3.6 Flash, and Gemini 3.5 Flash-Lite, delivering significant improvements in both cost efficiency and performance for video analysis tasks.3

3

How Agentic Video Understanding Works

Unlike traditional static video processing that ingests video at a fixed 1 FPS rate, agentic video understanding enables Gemini to take an active, goal-directed approach to video analysis.2

2

The system dynamically determines what to watch, how quickly to examine it, and which modality to use—whether visual frames, audio, or transcripts.3

3

This AI-driven video analysis approach allows Gemini to fetch only the moments and signals needed for each specific task, rather than processing entire videos uniformly.2

2

Previously, users could upload videos to Gemini, but the AI performed only static processing by splitting videos into individual frames.1

1

This resulted in slower performance and higher costs, particularly for long-form video analysis. With agentic video understanding, Gemini handles the entire workflow through an internal agentic loop, reducing the development overhead that developers previously faced when building such processes manually.3

3

Dramatic Cost Reductions and Performance Gains

Across standard video analysis benchmarks, 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%.2

2

3

3

These efficiency gains prove especially valuable for long-form video content, from 10-minute how-to guides to 90-minute lectures and multi-hour recordings, where static processing previously forced developers to choose between high token costs or techniques that dropped critical details.2

2

Google states that Gemini 3.7 Flash with agentic video understanding offers the best overall quality and the optimal combination of quality and cost efficiency among tested models, positioning it at the accuracy-to-cost Pareto frontier for video analysis.3

3

The feature uses standard Gemini API token pricing with no additional fees.3

3

New Capabilities Unlocked

Agentic video understanding unlocks several advanced capabilities for dynamic video analysis. The system enables sub-second moment retrieval, pinpointing split-second state changes and tight cut boundaries that static processing at 1 FPS would miss, making precise automated video editing possible.3

3

For long-form needle-in-a-haystack searches, Gemini can now answer complex queries across multi-hour videos without consuming millions of tokens.2

2

The feature also enhances anomaly detection by resampling interesting time windows at higher FPS to inspect rapid motion and subtle visual artifacts.3

3

Additionally, Gemini can now count and track repeated physical movements and distinct objects over time, providing precise object counting capabilities.1

1

3

3

Availability and Future Rollout

Agentic video understanding is currently available through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform for both video uploads and YouTube videos.2

2

3

3

Developers can enable the feature by setting processing to "agentic" in the API configuration.3

3

Google plans to roll out agentic video understanding to all users in the Gemini app across Flash and Flash-Lite models soon.1

1

3

3

In the coming months, the feature will also power Ask YouTube on the video watch page, using Gemini to provide answers grounded in video visuals.3

3

This could significantly benefit content creators who want to analyze their videos more efficiently with Gemini.1

1