Everything Google announced at I/O 2026: Gemini 3.5, Omni, Spark, and the Search that’s changed forever

If you had "agents in everything" on your I/O 2026 bingo card, congratulations. Google's keynote in Mountain View on Tuesday was less about new features and more about a quiet rewiring of every product the company owns. Sundar Pichai called it "hyper progress," which is the kind of phrase CEOs use when they want you to lean in. But the numbers do some of the lifting for him. Gemini is now at 900 million monthly users, more than double last year's count. AI Mode in Search has crossed a billion. People are using these tools, and Google has decided that the next move is to let those tools start doing things on their own.

Plenty to unpack, so let's get into it.

Gemini 3.5 Flash is fast, cheaper, and now the default everywhere

Gemini 3.5 Flash is the new default for AI Mode and the Gemini app. The pitch is that it outperforms Gemini 3.1 Pro on agentic and coding benchmarks while running roughly four times faster than other frontier models, at less than half the cost. Whether or not you care about Terminal-Bench scores, the practical effect is that Search and the Gemini app should feel snappier from today.

Gemini 3.5 Pro is still in internal testing and arrives next month. Google is using 3.5 Flash internally to power Antigravity 2.0—its agent orchestration platform that got its own moment in the keynote, complete with a developer demo that built a custom OS to run Doom. Antigravity is for developers, but it's worth knowing the name because it shows up everywhere else too.

Gemini Omni is the video model that actually listens

Omni is Google's new family of multimodal generation models, and Omni Flash is the first of them. Feed it text, an image, a video, audio, or some combination, and it generates editable video. The standout trick is conversational editing. You point at a metal sculpture in your clip and ask Omni to make it out of bubbles, and it does, while keeping the person walking through it consistent. Physics behave. Characters don't morph between shots.

There's also an avatar feature that lets you drop your own face into generated clips, with your own voice. Every output carries a SynthID watermark. Omni Flash rolls out today to AI Plus, Pro, and Ultra subscribers through the Gemini app, Google Flow, and YouTube Shorts.

Gemini Spark is Google's pitch for a 24/7 personal agent

Spark is the agent Google really wants you to think about. It's a personal assistant powered by 3.5 Flash that runs on virtual machines in Google Cloud—which means it keeps working when your laptop is shut. It connects to Gmail, Docs, Drive, and a growing list of third-party tools over MCP, including Canva, OpenTable, and Instacart.

The demos were the most specific part of the keynote. Spark can watch your credit card statements for sneaky subscription charges, pull project notes scattered across Gmail into a single Doc, or plan a block party with an RSVP tracker in Sheets and a hype deck in Slides. Anything involving spending money or sending an email triggers a confirmation prompt first.

Trusted testers get it this week. AI Ultra subscribers in the US get the beta next week. A macOS version that works with local files is planned for summer.

The Gemini app gets a new look called Neural Expressive

The app has been redesigned with what Google is calling Neural Expressive—new typography, brighter colours, haptic feedback, fluid animations. Gemini Live is now inline rather than a separate fullscreen mode, so you don't have to switch contexts to talk. Regional dialects are on the way. Responses now show up with inline images, narrated videos, and interactive timelines instead of just walls of text.

Daily Brief is the morning agent

Daily Brief is a smaller agent that sits inside the Gemini app and assembles a digest from your Gmail, Calendar, and Tasks every morning. It prioritises, summarises, and suggests next steps. AI Plus, Pro, and Ultra subscribers in the US get it starting today. It's the kind of feature that either becomes the first thing you check in the morning or quietly never gets enabled. We'll see.

Google Search just got its biggest box redesign in 25 years

This is the change that'll touch the most people. The Search box now expands as you type, suggests completions that go beyond autocomplete, and accepts text, images, files, videos, or even Chrome tabs as inputs. It's rolling out globally in every language where AI Mode is available.

You can also ask a follow-up question inside an AI Overview and slide into AI Mode without leaving the page. The deeper you dig, the more tailored the supporting links become. Personal Intelligence—which lets Gemini pull context from Gmail and Photos—is now free in 98 languages across nearly 200 countries.

Search agents and mini apps are coming this summer

This is where Search starts behaving less like a search engine. Information agents run quietly in the background, monitoring whatever you've asked them to track—apartment listings, sneaker drops, biotech stocks, news on a niche topic. They send synthesised updates and let you act on them. AI Pro and Ultra subscribers get them first.

Search is also gaining generative UI through Antigravity, which means it can build interactive layouts, simulations, and graphs on the fly when a static answer won't cut it. For recurring tasks like wedding planning or fitness tracking, it can build mini apps and dashboards that pull in live data. Generative UI is free for everyone this summer. Mini apps start with AI Pro and Ultra users in the US.

There's also agentic booking. Ask for a private karaoke room that serves food after 11pm on a Friday, and Search surfaces options with direct booking links. For categories like home repair, beauty, or pet care, Google can call local businesses on your behalf. Not a feature for everyone, but useful if you hate making phone calls.

Universal Cart wants to swallow your online shopping

Universal Cart is a Gemini-powered shopping cart that follows you across Search, Gemini, YouTube, and Gmail. Add things from anywhere, and the cart tracks prices, flags restocks, and surfaces deals. The PC-builder example Google used was a smart one—the cart spots when two parts won't work together and suggests swaps, which is the kind of mistake AI is genuinely good at catching.

Built on Google Wallet, the cart knows your payment methods, loyalty programs, and merchant offers. Checkout happens via Google Pay or by redirecting to the seller. US users get it across Search and the Gemini app this summer. YouTube and Gmail come later.

Ask YouTube drops keywords for conversation

YouTube Premium members in the US can now use Ask YouTube. Ask something like how to teach a three-year-old to ride a pedal bike or what cosy games to play in the evening, and YouTube assembles relevant videos across its catalogue, including Shorts, and jumps to the exact moment that answers your question. Follow-ups are supported. It rolls out more broadly later.

Gemini Omni is showing up inside YouTube Shorts Remix and the Create app too, where users can add themselves into existing Shorts or restyle them with an earlier-era aesthetic. Creators 18 and over get YouTube's likeness detection tool to manage how their face gets used. They can also opt out entirely.

Voice control comes to Gmail, Docs, and Keep

Google is rolling these out under the Live banner. Gmail Live searches your inbox with natural voice. "What's my flight gate?" "What's coming up at my kid's school?" It surfaces the answer instead of returning a list of email threads.

Docs Live is more impressive in motion. The keynote demo had someone verbally brain-dump a complex prompt, ask Gemini to pull a doc from Drive, then reformat the result on the fly. With permission, it pulls supporting details from Gmail, Drive, Chat, and the web. Keep Live does the same for rambling notes, turning them into structured lists.

AI Pro and Ultra subscribers get these features first. Workspace business customers get a preview this summer.

AI Inbox finally opens up

AI Inbox in Gmail was previously locked to AI Ultra subscribers and Workspace Enterprise Plus customers. It's now rolling out to AI Plus and Pro subscribers in the US, with personalised draft replies, instant access to relevant Docs, Sheets, and Slides, and one-click task management. Useful if you live in Gmail. Skip if you don't.

Google Pics enters the image editing race

Not Google Photos. Pics is a new Workspace app built on Nano Banana for generating and editing images. The selling point is precise control—you can click on one element in an image and edit just that. Swap a sweater colour. Turn a dog into a cat. Edit embedded text without losing the font. Shared canvases let multiple people edit the same image at once.

Trusted testers get it today. Wider access to AI Pro, Ultra, and Workspace business customers comes this summer.

AI plan prices got shuffled

Google is moving Gemini app users from daily prompt caps to a "compute-used" model. Simple prompts use less, complex video and coding tasks use more. Limits refresh every five hours until you hit a weekly cap, and AI Pro and Ultra subscribers can buy top-up credits.

The AI Ultra plan now starts at $100 a month, with five times the Gemini app and Antigravity usage of the $20 Pro plan, 20TB of storage, and a YouTube Premium individual plan thrown in. The old $250 Ultra tier dropped to $200—same capabilities as before, a 20x usage multiplier over Pro, and access to Project Genie.

SynthID and Content Credentials reach Search and Chrome

Google has watermarked over 100 billion images and videos and 60,000 years of audio with SynthID. Detection, previously only inside the Gemini app, is now expanding to Search through Lens, AI Mode, and Circle to Search, and to Chrome over the coming weeks. Ask "Is this AI generated?" and you'll get an answer, at least for content made with Google's tools.

C2PA Content Credentials are being added to verification too, so you can check whether an image came straight off a camera or was edited, and with what. Pixel 10 already supports Content Credentials for photos. Video support is coming to Pixel 8, 9, and 10 in the coming weeks. Authentic Pixel footage shared on Instagram will soon be labelled as such, since Meta is also adopting the standard. OpenAI, Kakao, and ElevenLabs are joining SynthID too.

Intelligent eyewear arrives this fall

Google insists on calling them intelligent eyewear, not smart glasses. The first set is audio-only, built with Samsung and Qualcomm internally and Gentle Monster and Warby Parker on the outside. They launch this fall in select markets. Prices later.

You can summon Gemini with "Hey Google" or a tap on the frame. Features include voice control, real-time translation that matches the speaker's tone, turn-by-turn navigation, photo and video capture, and multi-step task execution. The big stage demo had a presenter order a nitro cold brew through DoorDash without touching her phone—no hands, no unlock screen, just talking to her glasses. They pair with both Android and iPhone.

Project Aura, the display version Google is making with Xreal, also got an update. The compute puck has a new fingerprint sensor and a lanyard. Display glasses arrive after the audio set.

Android Halo and Wear OS 7 round out the platform

Android Halo is a UI layer that shows what your agent is doing at the top of your phone screen, so you don't have to keep opening apps to check. It launches later this year and works with Spark and other supported agents.

Wear OS 7 got a passing mention. It picks up Android 17's design updates, adds customisable Wear Widgets that mirror their phone counterparts, and supports the Create My Widget feature on the watch. A test version is on the emulator now.

Gemini for Science wants to speed up research

Gemini for Science bundles Google's research tools into one program with collaborators including Stanford, Imperial College London, and The Crick Institute. Three Labs prototypes handle hypothesis generation, computational discovery, and literature insights. Science Skills connect Antigravity to over 30 life science databases and tools like UniProt, AlphaFold Database, and AlphaGenome API—the kind of integration that matters if you're actually doing this work and is invisible if you're not.

Project Genie now builds worlds from Google Street View

Project Genie, Google's interactive 3D world generator, first showed up as a research preview in January. The update at I/O hooks it into Google Street View, so subscribers on the $200 Ultra plan can build worlds based on real US places and then mess with them. Cover a landscape in snow. Plunge a neighbourhood underwater. Generate a character to wander through it. Google's framing is that this lets its models "anchor themselves in reality"—corporate-speak for giving AI agents and robots a more realistic playground to learn in. Genie stays off the $100 Ultra tier for now.

Google Beam is experimenting with AI agents

Google Beam—the video calling system formerly known as Project Starline—is being tested with lifelike AI agents. The system is being demoed to attendees, with agents that can answer questions, read documents held up to the camera, and look up restaurant recommendations mid-call. The agents don't really appear in 3D the way a real person would on the $25,000 HP Dimension system, but the experiment hints at where Beam is going. Group calls are also in development and will work with Google Meet and Zoom.