I called it yesterday. Google I/O was going to be a big swing. They came out of the gate Monday and threw four punches in a row, so let me sort through what actually matters.
Gemini Omni Wants to Be the Video Layer of the Internet
Google announced Gemini Omni, a new model that takes any input and spits out video. Text, image, audio, whatever you feed it. They're rolling it out to Google AI Plus, Pro, and Ultra subscribers through the Gemini app and Google Flow this week, with YouTube Shorts getting it next week.
This matters because the video generation race just changed shape. OpenAI's Sora was the standalone moment. Google is distributing this through products people already use, in front of a billion-plus YouTube creators.
My take: pricing and quality will decide if this lands. The demos at keynotes always look gorgeous. The real test is what your aunt makes when she tries to remix a vacation video. If Omni handles that without producing the usual six-fingered nightmare fuel, this is a real product. If not, it's another demo reel.
Gemini 3.5 Flash and the Agentic Coding Pitch
Google also dropped Gemini 3.5 Flash, a faster model aimed at agentic coding and long-horizon tasks. They're claiming 4x faster than rival frontier models and major gains on coding benchmarks. Same announcement came with Antigravity 2.0, their agent-first coding platform.
This is the shot at Anthropic. Claude is the model my dev team reaches for when work has to ship. If Google can deliver speed plus a real agentic toolchain at a lower cost, the build versus buy math changes for a lot of shops.
My take: benchmark wins are cheap. What I want to know is how 3.5 Flash holds up at hour 4 of a real refactor, when context is sprawling and the model has to remember why it made a decision 200 turns ago. That's where Claude has been winning. Show me the long-tail performance and we can talk.
Gemini Spark and the $100 a Month Question
Google rolled out a new AI Ultra tier at $100 per month, aimed at developers, creators, and power users. The headline feature is Gemini Spark, which Google is calling a "24/7 AI agent." It's in beta, first to trusted testers and Ultra subscribers.
$100 a month is the same neighborhood as ChatGPT Pro and Claude Max. The pricing is now a triangle of three. The question is what you actually get for it.
My take: "24/7 AI agent" is marketing until I see the bill at the end of the month for actual usage. The Pro tier from OpenAI ran some heavy users a lot of compute, and there's only so long any of these labs can eat that. If Spark is genuinely always-on doing real work, the $100 is cheap. If it's metered behind the scenes and you hit a wall halfway through your project, it's expensive.
The Audio Glasses Are Coming
Google also showed off "intelligent eyewear" coming this fall. First product is audio glasses, partnering with Warby Parker and Gentle Monster on the frames. No display, just audio and an assistant.
My take: this is the smart move. Meta's Ray-Ban glasses have been the surprise hit of the last two years because they don't try to put a screen on your face. Audio plus a good assistant is the form factor that actually fits into how people live. The frame partnerships matter too. Warby Parker is the right look. Gentle Monster picks up the fashion-forward crowd. If the assistant is any good, these sell.
What I'm Watching Next
How fast does OpenAI counterpunch. Anthropic still hasn't closed that $30B round. The model release cadence is now essentially monthly across the three big labs, and the differentiation is moving from raw IQ to integration, distribution, and pricing.
Y'all stay sharp out there.