Gemini Launches Agentic Video Understanding in Flash Models
Developers gain the ability to dynamically search, scan, and inspect long videos in Gemini Flash models.
Rohan Doshi, Gemini multimodal PM at DeepMind, posted the launch of agentic video understanding in Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. The official Google account confirmed the update and shared a 58-second demo video plus a scatter plot on the 1H-VideoQA benchmark. The feature lets models scrub, zoom, edit, and resample frames instead of ingesting video statically. Posts note an attached chart comparing tokens per query and accuracy, with 3.7 Flash ranking at the frontier. LangChain replied that agents can now use the capability.
Combined views
524.8K
6 posts, first seen 13h ago
Reactions
6 postsGO 
Google@GoogleWe’re introducing a new capability to our latest Gemini models: agentic video understanding. This allows developers to process long-form video content with more accuracy, while using up to 88% less tokens. See how it works 🧵
- likes: 4.5K
- replies: 159
- bookmarks: 2K
- reposts: 349
GO Google@GoogleWith agentic video understanding, developers can dynamically search, scan, and inspect long-form video with up to: 📉 88% fewer tokens 📉 66% lower cost ✅ 7% better accuracy
- likes: 183
- replies: 6
- bookmarks: 24
- reposts: 8
RD Rohan Doshi@RohanLikesAIToday, we’re launching Agentic Video Understanding in Gemini across 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. 📉 ~88% fewer tokens 💰 ~66% lower cost 📈 ~7% higher accuracy The result: a new Pareto frontier for video understanding across accuracy and cost. Traditional video…
- likes: 218
- replies: 12
- bookmarks: 94
- reposts: 24
RD Rohan Doshi@RohanLikesAIToday, we’re launching Agentic Video Understanding in Gemini across 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. 📉 ~88% fewer tokens 💰 ~66% lower cost 📈 ~7% higher accuracy The result: a new Pareto frontier for video understanding across accuracy and cost. Traditional video…
- likes: 218
- replies: 12
- bookmarks: 94
- reposts: 24
👩💻P 👩💻 Paige Bailey@DynamicWebPaigeGiving Gemini the ability to dynamically scrub, zoom in, edit, and re-sample frame rates instead of mindlessly ingesting flat video changes everything for long-form video reasoning and editing! 📹🎞️ • 88% fewer tokens • 66% cost reduction • Sub-second precision and dynamic FPS…
- likes: 86
- replies: 2
- bookmarks: 21
- reposts: 4
SR Sydney Runkle@sydneyrunkleyou can now build @LangChain agents w/ agentic video understanding! google is consistently at the forefront of multimodal processing; this is the video analog of "agentic vision" token usage + cost is also way down w/ this new approach! docs:
- likes: 91
- replies: 4
- bookmarks: 58
- reposts: 14
