Google’s annual I/O conference dropped two models worth paying attention to: Gemini 3.5 Flash and a video-generation model called Gemini Omni.
Gemini 3.5 Flash is the first release from the Gemini 3.5 family, and it’s already live — available through the Gemini app and inside AI Mode in Google Search. Google positions it as matching flagship-level intelligence at Flash speeds, which is the series’s whole pitch. What’s different this time is the benchmark performance. It’s outscoring Gemini 3.1 Pro on coding and agentic tasks, which is a meaningful jump given that Pro is the higher-tier model in that family. It’s also taken the default model slot, replacing whatever was there before.
Gemini Omni is the more interesting announcement. Feed it images, audio, video, text — any combination — and it generates video output grounded in what Google calls “real-world knowledge.” The first model from this family, Gemini Omni Flash, lets you go back and change things after the fact through conversation, tweaking specific elements or overhauling entire scenes without losing continuity from the original. Google says the model has a better grasp of physical forces — gravity, fluid dynamics, kinetic energy — which should translate to more believable motion and scene behavior rather than the floaty, physics-ignoring output that’s plagued most video generators.
There’s also an Avatars feature that creates a digital version of you, and voice input is supported. Every video gets SynthID watermarking baked in, Google’s method for flagging AI-generated content.
Gemini Omni Flash is available today for Google AI Plus, Pro, and Ultra subscribers globally, accessible through the Gemini app and Google Flow. It’s also rolling out free on YouTube Shorts and YouTube Create.
