TL;DR — On 2026-09-01 Google launched agentic video understanding on Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. Instead of ingesting a video at a fixed 1 FPS, the model searches frames, audio, and transcripts for the moments that answer the question. Official benches: up to 88% less usage, 66% lower analysis cost, 7% higher accuracy. If you just need chapters from a YouTube lecture, paste the link — you do not have to set processing=agentic.
Paste a long-form video URL — BibiGPT turns it into chapters, a transcript, and follow-up Q&A.
As of 2026-09-04, Google launched agentic video understanding on 2026-09-01 across Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. The model searches frames, audio, and transcripts instead of reading the file at a fixed 1 FPS. Official benches: up to 88% less usage, 66% lower analysis cost, 7% higher accuracy. If you need a lecture turned into searchable notes, paste the link into BibiGPT instead of wiring processing=agentic.
A processing mode, not a new model name. The model decides what to watch, at what speed, and through which modality — frames, audio, or transcript — instead of reading every second at a fixed rate.
Google enabled it on Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. 3.7 Flash is the quality-and-cost frontier on Google's own benches. Developers turn it on by setting processing to agentic in the Gemini API.
Static mode ingests video at a fixed frame rate (default 1 FPS, adjustable). Agentic mode uses native video tools to scan, zoom, and inspect only the segments that answer the query — including sub-second cuts that 1 FPS would miss.
On standard video-analysis benches Google reports up to 88% less usage, 66% lower analysis cost, and 7% higher accuracy versus static processing. Gains are largest on long-form (10-minute how-tos to 90-minute lectures). Those are maxima, not a single-test guarantee.
The API mode helps teams that already run Gemini themselves. It does not replace a product that files chapters, a transcript, and follow-up Q&A for a human.
Paste a YouTube or lecture URL, get chapters and a transcript. The point of a video assistant is the artifact — not an API config flag.
A 90-minute lecture still needs timestamps and export. A cheaper agentic call you do not file is wasted; a chaptered summary is the file.
The 2026-09-01 launch is API + Google AI Studio + Gemini Enterprise Agent Platform, for uploads and YouTube URLs, at standard usage rates with no extra fee. Consumer Gemini App and YouTube Ask YouTube are listed as coming later — not this page's job to promise a date.
Headline shifts from Google's 2026-09-01 agentic video launch.
Agentic video understanding is a processing flag on existing Flash models. You do not buy a fourth Gemini. Developers set processing to agentic in the Gemini API.
Static mode samples at a fixed FPS (default 1). Agentic mode decides what to watch, at what speed, and whether to read frames, audio, or the transcript — fetching only the moments the query needs.
Google reports up to 88% less usage, 66% lower analysis cost, and 7% higher accuracy versus static processing. Long-form benches (1H-VideoQA, LVBench) show the 88% usage drop; harder reasoning benches save less.
Video uploads and YouTube URLs via the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform. Standard usage rates, no extra fee. Gemini App and YouTube Ask YouTube are later surfaces.
Gemini 3.8 Flash (2026-09-02) lists agentic video understanding as Preview. This page is the 09-01 capability launch. One intent, one URL — do not fork a second explainer for the same search.
Where the API mode matters — and where a summarizer is the actual product.
Turn on agentic processing for long-video agents so you stop paying for every frame. Then still dump the finished video into BibiGPT so humans get chapters instead of a raw usage dump.
A 90-minute course video does not need an API flag. Paste the URL, export the transcript, ask follow-up questions. That is the BibiGPT path.
Omni is generate-from-any-input. Agentic video understanding is analyze-the-file-you-already-have. Keep the Omni explainer for creation; this page is the analysis-mode URL.
Launch claims come from Google's own posts and the Gemini API pricing page.
On 2026-09-01 Google launched agentic video understanding on Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite, reporting up to 88% less usage, 66% lower analysis cost, and 7% higher accuracy versus static processing.
Google — Introducing agentic video understanding with Gemini ↗Gemini API pricing notes (updated 2026-09-03): agentic video understanding bills at standard usage rates; usage usage is variable based on content loaded rather than full video length.
Google AI for Developers — Gemini API pricing ↗Agentic video understanding is a Gemini API processing mode launched on 2026-09-01. The model searches visual frames, audio, and transcripts for the moments that answer a question, instead of ingesting the whole file at a fixed 1 frame per second.
Why people use BibiGPT to turn videos into text every day.
Trusted by 50,000+ users worldwide
“I paste a link and get clean captions in seconds — it saves me hours of retyping every single week.”
Maya R.
Content Creator · Repurposes short videos
“Exporting the transcript lets me review new words at my own pace instead of pausing the video constantly.”
Daniel K.
Language Learner · Studies with real videos
“Accurate, timestamped text I can quote directly. It has quietly become part of my daily workflow.”
Priya S.
Researcher · Cites public talks
FAQ'S
Ask us anything!
1 Best Bilibili AI video summary tool 2026? Paste a link for a free AI recap, mind map, and transcript-style takeaways on 30+ platforms — no login needed.
2 Install a video-summary skill into DeepSeek Harness in one copy-paste — SKILL.md matches Claude Code. Or skip dsh: paste a Bilibili or YouTube link in the browser and get a timestamped summary.
3 Reverse any video free with no app and no download, right in your browser. Step-by-step for iPhone, Android, and PC, plus how to reverse audio (2026).
Paste a long YouTube, lecture, or podcast URL into BibiGPT. You get chapters, a transcript, and follow-up Q&A — without setting processing=agentic or picking a Flash SKU.
