Google ran its whole keynote on one idea... AI should stop answering and start doing. Two new models, a fleet of always-on agents, a free platform for building your own, smart glasses, and a quiet price war on its top plan. Here's all of it. What each one does, what it costs, when you get it, and what's worth your attention.
If you read nothing else, read these six lines.
Gemini Omni makes and edits video from text, images, audio, or video. It's free inside YouTube Shorts.
Gemini 3.5 Flash puts frontier-grade AI at Flash prices. Frontier capability just got cheap.
Gemini Spark is a 24/7 agent that works while your laptop is closed. It's also $100 a month, US-only.
Antigravity 2.0 lets anyone run a whole team of AI agents on a project. Free, today.
Search is being rebuilt for conversations and background agents, not keywords.
A new $100 AI Ultra tier appeared. The old $250 plan quietly dropped to $200.
The eight launches that matter most. Features, pricing, release dates, and what Google actually demoed.
Google's "create anything from any input" model. Feed it text, images, audio, and video, and it builds one coherent video. Then you edit that video by chatting.
Google's demo... a claymation protein-folding explainer built from one prompt. Early tests put it behind Seedance 2 on cinematic feel, ahead on prompt control.
Google's newest model. Frontier-grade intelligence at Flash speed and Flash price, built to act, not just answer.
Already in use at Shopify, Macquarie Bank, Salesforce, Ramp, Xero, and Databricks.
A 24/7 cloud-based AI agent in the Gemini app. It does real work for you while your laptop is closed.
Google's demo... planning a block party. An RSVP spreadsheet, a doc tracking who brings what, reminder emails sent on their own.
Google Labs' AI design tool, now a real-time partner. Describe a UI by voice or text and watch it build live.
A standalone, agent-first desktop app. A command center for running a whole team of AI agents on one project.
For a limited time, AI Ultra subscribers get $100 in bonus Antigravity credits. Claim it in-app. The offer ends May 25, 2026.
The biggest Search-box redesign in 25 years, plus AI agents that watch the web for you around the clock.
Google's demo... finding a private karaoke room for six on a Friday that serves food late, with booking links pulled together for you.
One intelligent shopping cart that follows you across Search, Gemini, YouTube, and Gmail, hunting deals in the background.
Launch merchants include Nike, Sephora, Target, Ulta, Walmart, Wayfair, and Shopify stores.
Google's intelligent eyewear. Gemini in your ear now, and in your lens later, built with Samsung, Qualcomm, Gentle Monster, and Warby Parker.
Designer frames from Gentle Monster and Warby Parker. Display glasses were shown, but given no firm date.
Fourteen more launches that didn't headline but still move things.
The larger "orchestrator" model. It plans the work, then hands sub-tasks to Flash. In testing now.
Now builds native Android apps in production Kotlin, deploys to Cloud Run in one click, and exports straight to Antigravity.
Spin up a fully provisioned agent with one API call. Reasoning, tools, and code execution included.
An open standard that lets websites expose functions directly to AI agents. The plumbing of the "agentic web."
A personalized morning digest in the Gemini app. It pulls Gmail, Calendar, and chats into one action plan.
An AI image tool built on Nano Banana. Select and edit single objects, rewrite text inside an image, co-edit on a shared canvas. It's not Google Photos.
Omni-powered video creation, plus a Lyria 3 Pro music tool with section-by-section editing. Now in 140+ countries.
Conversational video search. Ask a complex question, get an interactive set of the most relevant clips.
AI-content watermarking and provenance, now verifiable in Search and Chrome. 100B+ images and videos already marked.
A status-bar layer that shows, at a glance, what your agent is working on right now.
A new Gemini-first laptop category, the Chromebook's successor, with a Magic Pointer and custom widgets.
A watch refresh. New Wear Widgets in a grid replace the old fullscreen Tiles.
A full redesign of the Gemini app. Fluid motion, inline Gemini Live, richer visual responses.
Conversational voice features. Search your inbox or co-write a doc by talking to it.
Only one real price story this year. A new $100 Ultra tier appeared, and the old $250 Ultra dropped to $200. Here's the full picture.
Google has a habit of announcing before shipping. Here's what's actually usable today versus what's a promise.
For each of these, the Default is what any AI would tell anyone who asked. The Expert read is what changes once you apply domain knowledge, reframe the question, or notice what a generalist would miss. The gap is the whole point.
"Google launched a powerful new AI video model. Use Gemini Omni to generate video content for your brand."
Omni's generation quality isn't the headline. Early hands-on tests rank it behind Seedance 2 on cinematic feel. Three things matter more for ad work. First, conversational editing, which removes an object or rewrites a scene without re-rolling the whole clip. Competitors can't do that. Second, accurate text rendering, which Google's own DeepMind lead called "really useful for advertising," where product names and slogans have to be right. Third, distribution, since it's free inside YouTube Shorts. The avatar feature also means talent-led UGC with no shoot and no creator fee. Treat Omni as an editing and iteration layer, not a Veo replacement.
"Gemini 3.5 Flash is faster and smarter. Switch to it for better, quicker results."
The speed is getting the attention. The price is the real story. Frontier-grade capability just collapsed to Flash pricing, roughly a third to half the cost of comparable models. Google's claim of $1B a year in enterprise savings is the signal. This is a pricing event dressed up as a model launch. Every content pipeline, research agent, or multi-step automation you shelved last quarter for being too expensive is worth re-costing this week. The constraint that killed those projects just moved.
"Gemini Spark is a 24/7 AI agent that works for you. Sign up to get an assistant that runs while you sleep."
You can't sign up. Spark is Ultra-tier at $100 a month, US-only, in beta the week of May 26. Waiting on the product is the wrong move. The reusable part is the architecture... cloud-VM agents running on the Antigravity harness. Antigravity 2.0, its CLI, and its SDK are free and shipped today. If you want a 24/7 agent, build the pattern now with the free tools. The capability is open. Only the consumer wrapper is locked.
"Stitch lets you design interfaces with AI. Use it to mock up your ideas faster."
For a marketer, Stitch is a same-day path from idea to a live, shareable asset. It streams a real design as you steer it by voice, generates a shareable link, and publishes to the web through Netlify. AI Studio deploys full apps to Cloud Run, with the first two free. Landing pages, quiz funnels, lead magnets, interactive offers... concept to live URL, with no developer and no Webflow subscription. You can test an offer the same afternoon you think of it.
"Google added AI agents to Search and shopping. Keep your SEO and product listings current."
Three announcements point to one structural shift. The web is being rebuilt for agents, not just humans. The new Search box is built for long conversational queries, and 3.5 Flash is the AI Mode default. Being the answer an AI puts together now beats ranking for a keyword. WebMCP lets sites expose functions directly to agents, and Expedia, Shopify, Instacart, and Target are already testing it. Universal Cart means an AI, not a person, increasingly does the buying. Within roughly twelve months, "is my site and catalog agent-readable?" becomes a real question, the way "is it mobile-ready?" was in 2012. Audit clients for AI-answer presence and structured-data readiness now.
"Google's SynthID watermarks AI content so people can spot what's fake. Good for trust."
If you ship AI content for clients, this is now an operational fact, not a curiosity. Every Omni video carries a SynthID watermark, verifiable in Search and Chrome with one click. OpenAI, ElevenLabs, and Meta are all on the standard. That has two consequences. Undisclosed AI is becoming technically undeniable, so bake disclosure into your process before a platform forces it. And provenance becomes a selling point. "Verifiably real," or "verifiably ours," is a positioning angle for any brand worried about deepfakes of its founder or talent.