Introducing agentic video understanding with Gemini
If you or your team analyse video content with Gemini - training footage, recorded meetings, site inspections - this update makes that far cheaper and more accurate, so it is worth revisiting any video workflows you had previously ruled out as too costly.
What changed
Source summary
Google launched agentic video understanding across the Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite models, using dynamic video scanning that cuts token usage by up to 88%, reduces cost by up to 66%, and improves accuracy by up to 7% compared with the previous static video-processing approach.
Why it matters to your work
Practical interpretation
If you or your team analyse video content with Gemini - training footage, recorded meetings, site inspections - this update makes that far cheaper and more accurate, so it is worth revisiting any video workflows you had previously ruled out as too costly.
When this could help
Worth enabling if you analyse video content via the Gemini API and want to reduce costs and improve accuracy, particularly for long videos where fixed-framerate processing is wasteful.
Agentic video understanding lets Gemini actively decide what to watch, at what speed, and through which modality (frames, audio, transcript) instead of processing every frame. This delivers up to 88% lower token consumption, 66% lower analysis costs, and 7% higher accuracy — especially on long videos.
How to get started
In the Gemini API, set processing to 'agentic' in the API configuration. The model will then dynamically choose which parts of the video to examine closely rather than processing every frame.
Provider guidance ↗
Try this · The SCALA method
Give AI a clearer brief
- Situation
- You have video content to analyse — finding specific moments, detecting anomalies, or extracting structured information.
- Consultant
- Act as a video analysis assistant using agentic video understanding.
- Action
- Analyse this video and [describe what you need: find the moment where X happens / count occurrences of Y / detect anomalies in Z]. [Fill in your actual analysis task.]
- Language
- Structured results with timestamps and confidence indicators.
- Ask
- For long videos, start with a specific question rather than 'summarise everything' — the agentic approach works best when it knows what to look for.
SCALA suggested prompt · Adapt the context and check the result.
Learn the SCALA method →Build it into your routine
SCALA suggested workflow
Cost-effective video analysis
For any video analysis task via the API, enable agentic processing to let the model skip irrelevant sections and focus on what matters — especially valuable for multi-hour recordings.
What to check
Sources & provider guides
Official documentation checked 12 Oct 2026. SCALA prompts and workflows are suggestions, not claims of hands-on testing.
Source & availability
This is a summary of a published provider announcement, with practical context from SCALA’s feed.
Plan, region, workspace and rollout conditions may affect access in Australia. Check the original source for current availability.
blog.google