Gemini Flash cuts video analysis costs by 88%

Google just gave its Gemini AI a sharper eye for videos and a lighter wallet to go with it. The tech giant said today that its latest Flash models—Gemini 3.7, 3.6, and 3.5 Flash-Lite—can now analyze uploaded videos far more accurately, while cutting the computing cost by up to 88 percent. Until now, users could drop videos into Gemini, but the AI only processed them frame-by-frame, a slow and expensive approach.

The upgrade is part of Google’s broader push to make Gemini smarter and cheaper. Earlier this year the company released new versions like Gemini 3.7 Flash, and teased more mobile features. The new “agentic video understanding” lets Gemini decide what to watch and when, picking the best mix of frames, audio, and transcripts. That means it can spot a blink-and-you-miss-it change, answer questions across hours-long videos, hunt for visual glitches, and even count objects or track movements in real time.

So far the feature is only live in the Gemini API inside Google AI Studio and the Enterprise Agent Platform. Google says it will reach the main Gemini mobile app “soon.” The company also plans to plug this tech into YouTube’s “Ask YouTube” tool, giving creators a way to query their own footage without burning through credits.

Fewer tokens, better accuracy, and deeper video skills could help Google close the gap with rivals like OpenAI’s video-analysis tools. For everyday users, it means fewer bill shocks and faster answers. For developers, it’s a cheaper route to build apps that need eyes on videos—everything from classroom tools to security bots.