Google I/O 2026: Live updates on XR glasses, Gemini, Search and more

The show's over, but if you'd like a deeper look into Google I/O, you can find links to all of Engadget's coverage below:

• Everything announced at Google I/O 2026

• Samsung and Google just teased their upcoming Android XR smartglasses at Google I/O

• Project Genie adds Google Street View integration and goes live for global AI Ultra users

• Google's Universal Cart will keep an eye on your shopping

• Google brings more conversational features to Gmail, Docs and Keep

• Google says Gemini 3.5 Flash rivals 'large flagship models' for coding and agentic tasks

• Google's Gemini Omni can generate 'anything from any input,' starting with video

• Google's newest app is an AI-powered image editor

• Google's redesigned Gemini comes with a new interface and AI models

• Ask YouTube compiles video answers to your questions

• Google's Gemini Spark is an agentic AI assistant

• Google shoves more AI into Search, including a dynamic Search box and agentic features

• Google announces Wear OS 7
  

If you joined us for the liveblog, thanks for joining us :)

You heard it here first! With that, we are finally wrapped, thanks for tuning in.

And without much fanfare, that's the end of the keynote.

Demis just said when we look back at this time we'll realize "we were standing in the foothills of the singularity." Lol, Lmao even.

Demis Hassabis is back to talk about Gemini for Science and WeatherNext, a weather forecasting model that Google designed to predict hurricanes.

A look at how Gemini/Nano Banana/WearOS can work together with these glasses. Not sure this is the kind of thing most people are looking for in smart glasses, but it's a nice flex for Google.

We're getting a demo of Nano Banana integration and smartwatch connectivity with a twist on the annual Google I/O crowd selfie. The presenter asked their glasses to snap a photo of the audience and add a blimp in the background with the I/O logo.

Google Maps integration with walking navigation is definitely intriguing. Meta's non-display glasses don't have navigation abilities, and the version on its display glasses can be hit and miss in my experience. 

Here we go, the first two designs for the Gemini-enabled smart glasses (no display). 

Google is sharing a sneak preview from Samsung. The two frames look very similar to Meta's AI Glasses, with clearly visible camera lenses.

We're going to have to wait a while still for display-enabled glasses, but Google is revealing its audio-only frames, which sound like their answer to Meta's Ray-Ban lineup.

Google is also working with Samsung on this project. They'll work with both Android and iOS device.

Google will release its first Audio Glasses this fall. The company is working with Warby Parker and Gentle Monster on these frames.

Folks, I know this is getting long in the tooth, but we're getting an update on Android XR. That's something, right?

Really weird demo of an app Google is calling Google Flow Music, which allows users to upload a recording and then prompt Gemini 3.5 to generate additional musical elements.

I'm not a filmmaker, are these good?

There have been more than a few moments where the presenters have awkwardly waited for the audience to clap.

You can use Pics to upload existing images to Workspace and edit them. For example, it's possible to remove and resize both foreground and background elements. Each generation is fingerprinted with Google's SynthID watermark.

I do appreciate AI-powered photo editing, though it feels very much like table stakes for consumer AI. 

Suz Chambers, director of Google Creative Lab, takes the stage to announce Google Pics. A new image generation tool inside of Workspace.

Another little feel-good video interlude where someone says that the "power of AI" is all about giving workers dignity. Okay then.

A little preview of the new multimodal capabilities coming to macOS with the previously mentioned dog-boarding inquiry.

In one of the more nifty demos we've seen so far, Woodward uses the Gemini app on macOS to write an email to a dog daycare. What's cool here is he prompts Gemini to write a set of tables using information from a set of documents he highlighted in Finder.

Woodward promised three product updates, but lucky us we're getting a bonus one related to the Gemini app on macOS.

Of course, AI agents are coming to the app as well. The first one is called Daily Brief, and as you can imagine, it pulls from your Google services to build an overview of your day. Power users will be able to build their own agentic workflows.

If you're a paying subscriber, you'll also get access to Gemini Omni inside the app starting today too. Woodward calls this the Nano Banana moment for video generation.

The Gemini app is getting a "stunning" new design Google calls "Neural Expressive." It uh, looks like they added some new gradients? 

This update will arrive in the summer... No, it's available starting today.

Google has refreshed the app using a new design language it calls Neural Expressive (a play on Android's Material Expressive language). Plus, the new app has support for additional regional dialects.

We're getting a recap of all the new features Google has released for the Gemini App over the last year.

Popping in to say I'm praying for everyone's bladders. Can Gemini do that?

After a brief interlude, Josh Woodward is back to chat about the Gemini App.

I know I'm starting to sound like a broken record, but the Universal Cart will arrive in the US this summer.

One of the those new features is something Google calls the Universal Cart. It's a cart that follows you across different Google services. It can notify you when items go on sale or are back in stock. In a very relatable example, the cart can also notify if it notices different products that won't work together, such as a PC motherboard and processor.

Would you trust an AI agent with your credit card and a shopping list? That's basically what Google is promising right now (I, uh, have a lot of questions.)

In any case, Google is now talking about the tech it's building to make agentic commerce possible, including a new cross-industry protocol.

Had a brief Wi-Fi hiccup there, folks.

We still have a little over an hour to go, folks. Here in Mountain View, the temps are quickly changing from "pleasant" to "we're going to have a bad time as soon as we leave the shade."

We're now getting a demo of the code generation. For the purpose of this demo, we get to see Gemini 3.5 Flash code in real time.

I am also very much not a visual learner. I would much rather Google show me (preferably non-AI-generated) text than a cute little graphic.

Thanks to Gemini 3.5 Flash and Antigravity integration, AI Mode can build custom interface elements for user questions. Google plans to roll out this capability, you guessed it, this summer. It will be free for everyone.

Do you understand black holes now?? (I don't.) 

And this is where I wonder if Google has lost the plot a bit.. who is asking for agentic coding in search? 

Robby Stein, VP of product for Google Search, is now on stage to announce that coding capabilities, courtesy of Antigravity, is coming to AI Mode.

These agents will work with Gemini Spark and they can continuously monitor the web for updates to your questions. Google plans to debut them in the summer, followed by other, more specialized agents down the road.

So all of this is basically turning agents loose inside search. There are some interesting possibilities there, but I'm also wondering about all the ways this could go wrong (remember when AI answers said glue was a good pizza topping?)

Surprise! AI agents are coming to search too, which you can task with complex queries.

So this is basically the equivalent of Google's smart replies but inside the search box. As someone who often has very specific search queries, I'm not sure how I feel about Google's AI trying to guess what I'm looking for.

Google is launching a new search box. Reid says this new interface goes beyond autocomplete, with AI-generated suggestions. It also natively multi-modal, allowing you to upload photos and videos as part of your queries. Google is launching this today.

As you might expect, AI Mode is getting a Gemini 3.5 Flash upgrade.

Anyway, now we got Liz Reid, VP of Search, to talk about AI Mode and AI Overviews.

Did Pichai just call Google Search its ultimate moonshot?

Agentic browsing is one of those things I feel like we've been promised for awhile, but hasn't really happened. Sounds like that may change with Gemini Spark, which is coming to Chrome later this year. 

Later this year, Google plans to release a dedicated home for all of the AI agents it offers on Android called Android Halo.

On that note, there's a new AI Ultra tier that starts at $100, and the existing $250 plan is dropping down to $200.

As you imagine, Spark connects to all of Google's other services, so it can connect to Sheets, Gmail and other apps to coordinate tasks across them all. Google plans to roll out Spark to AI Ultra subscribers in the US.

Sundar references another AI meme, saying that when using agent platform Gemini Spark "yes, you can close your laptop." (This was recently the source of some very low-stakes conflict in my house when I closed my husband's laptop while Claude was running...) 

It's a 24/7 remote agent, meaning it can run tasks even when you're away from the computer.

Pichai is now back to announce Gemini Spark, a new personal AI agent. He gives the stage to Josh Woodward to demo Spark.

On that note, Google is rolling out Gemini 3.5 Flash today across all of its products and APIs. As for 3.5 Pro, it will arrive sometime next month.

If you want to try the new Antigravity for yourself, it's available to download today.

Ah yes, the ultimate test for any operating system: Can you play Doom on it?

Next, we get a demo of a Doom clone Antigravity coded for the OS the system built previously.

Mohan says Google tasked Antigravity 2.0 (and its army of agents) with building an operating system from scratch. It did this using less than a $1,000 of tokens.

Google is clearly trying to catch up to Claude Code's dominance. I have to say Antigravity is a good name. 

Varun Mohan, the head of Google's Antigravity platform, is now on stage to talk about the company's agentic coding platform. Not much to say here if you aren't a developer, but Mohan says the latest version of the software is "unabashedly agent-first."

Now we're onto Gemini 3.5, Google's newest flagship model. Pichai, who's back on stage, says it's four times faster than other frontier models.

SynthID and content credentials are not the flashiest updates Google will share today, but as AI models evolve it's incredibly important for people to have tools to understand what is AI generated and what isn't. Sundar uses this meme as an example of Gemini's new AI detection capabilities. 

The model offers conversational editing, allowing you to change elements like characters, background and more with your voice. According to Hassabis, down the road, Omni will be able to generate any output from any input. The first model in the family, Gemini Omni Flash, arrives this summer.

These videos always look very slick and impressive, but the real test will be how well it works IRL. Anyone remember the difference between last year's Veo demos and the initial clips made by real users? That said, the multimodal capabilities of Omni is definitely intriguing.

Gemini Omni can generate "anything from any input."

Hassabis says artificial general intelligence is just a few years a way (a claim he loves to repeat). He announces Gemini Omni, a new video-generation model that builds on the company's expertise in world models.

Demis Hassabis, the CEO of Deepmind, is now on stage to talk about world models.

Of Google's latest 8i TPU processor, Pichai says it allows Google to distribute its training across the world. It's also faster and more energy efficient.

This might be the earliest in an I/O keynote we've gotten a TPU deep dive. Another sign of the times, I guess.

Docs Live (and versions coming to Gmail and Keep) certainly seems like something that could be very handy for those with a variety of disabilities! Being able to tell Docs to move sections "up top" and format them with bold or italics instead of having to do so with your fingers could be much easier.

Cher, you mentioned accessibility earlier, and Docs Live seems like something that could have some interesting use cases there. I know a few folks who swear by Docs' voice typing feature and Docs Live would be next level. 

We're getting a demo now of Docs Live, which allows you to use your voice to generate documents. It's rolling out this summer too.

Soon, you can watch even less of a YouTube video :) Ask YouTube is starting to test now. (I joke but this is actually pretty useful.)

Pichai takes a moment to speak about Ask YouTube, noting the company plans to roll out the feature throughout the US this summer.

A lot of that growth has come courtesy of the company's Nano Banana image generator.

Pichai now turns to AI Mode and the Gemini app. He says the latter has more than 900 million monthly users.

There's a token leaderboard for you. (I do wonder what other companies' numbers here would look like...)

I was wondering how soon we would hear about tokens. Pichai says Google is processing 3.2 quadrillion tokens per month, a 7x year-over-year growth. You read that right.

He begins with a recap of Google's past 12 months, calling it a year of "hyper progress." But first we get a AI-assisted gag of everything Pichai has done recently, including learning to play guitar.

Sundar says he has had an "intense year." I imagine every AI CEO feels that way at this point.

Naturally, first to take the stage is Google CEO Sundar Pichai.

Are we feeling inspired yet, folks? 

Okay, we're starting with an AI-generated countdown.

No more AI-generated music. Instead, you got your usual tech event playlist of recent indie hits, which is how you know things are getting serious.

Shoutout to the crew currently mopping the stage. Can't have the execs coming out to a not perfectly-shiny floor, I guess.

We're in a brief intermission as the Shoreline crew readies the stage for the keynote.

I'm pretty impressed at all the people playing this game right now but anyway, here's your 9 minute warning. Grab your snacks and gird your loins...

Okay, game time is (mercifully) over. 

Co-host and "Pro" gamer CouRageJD is sweating to try to get on the game's leaderboard. Despite his best efforts, he gets nowhere close. Someone in the US managed to climb more than 800 levels.

Good? It's giving Subway Surfer (derogatory) to me.

Also, shoutout to Karissa who recently used the term "AI slopidemic" in one of her articles. I need to find a way to incorporate that into one of my pieces.

I do not hate these levels? If there's anything AI is good for... This seems like a good scenario for it.

I know a friend of a friend who has performed on this very stage before. What can I say? I know people.

Right now, we're looking at some of the AI slop levels the audience helped generate.

I never thought I'd say it but I kind of miss the chaotic HUMAN DJ sets that we're so used to seeing at this point. I guess everything has to be an AI flex now.

Cher, Ray says he loves you too <3

You can't see me readers, but there's a single tear running down my cheek thinking about the data center compute Google is burning through running this game right now.

"Friend of Engadget"? Ray Wong? Hmm... 

(JK Hi Ray)

Looks like I just stumbled onto this gameplay session. The background music is a vibe though. Seems people can "join the game" by going to Google's Infinite Scaler website.

Gizmodo consumer tech editor and friend of Engadget Raymond Wong complains the game isn't loading for him.

They said levels can include anything you want as long as it's safe for work. 

Of course, just as I say that, the pre-show has started and the two hosts are introducing a game called Infinite Scaler. The audience needs to help generate levels for the game using prompts they write on their phones.

Alright everyone, T-minus 28 minutes till the keynote begins... Has a pre-show started, Karissa and Igor?

I imagine Google has been planning the contents of this keynote for months, but I wonder if the company saw the backlash to the commencement speech former CEO Eric Schmidt gave recently where he was booed for what he had to say about AI, and if it decided, as a result, to change course.

I miss when they randomly launched hardware at I/O, a developer conference. Weren't there some years were developers were the first to get devices to test on? I can't recall, everything blurs together in my mind now...

I know. My first I/O was in 2017. It was the year Google introduced Digital Wellbeing. I miss those days. There were way more announcements you could get excited about.

AI, AI and more AI? Anyone remember the days when I/O was all about Android and they would throw in the AI stuff at the very end? Now, it's entirely about AI and Android is relegated to an online only event a few days before.

While we wait for the pre-show to start. Any predictions for what we will see Google announce today?

The Shoreline Amphitheatre is slowing filling up.

Is a Four Tet like a Four Loko or.. The irony of me just going to Google for the answer as I liveblog a Google event... Is not lost on me.

What is a Four Tet Igor. Oh dear am I showing my utter lack of coolness here.

Cherlynn, I would describe the music as "what if Four Tet but jelly fish?"

I love jellyfish but it is kind of a bummer to see Google replace what is usually a human DJ with a generative AI music experiment. 

No live performance as of yet. In 2025, Google hired Toro y Moi to play. Any guesses who we might see this year?

Oh so would you say this music is... fishy..? Or jelly...

It's actually a pretty cool AI model powering the experience. It interprets a live feed of jelly fish from the Monterey Bay Aquarium, and as they move a grid, it alters the amount of bass, melody, drums or chords in the song mix.

Not much happening here yet, but there is some kind of AI-powered jellyfish visualization onscreen at the moment. 

By good seats for any such event, by the way, I mean a stable surface on which to place our laptops, nearby outlets in case we run out of power, strong and steady Wi-Fi connections and a bottle of water on hand. Some companies would never allow a power outlet to be visible, let alone have media be able to plug their laptops in during a keynote. Any guesses?

We just got our seats inside the venue. We're waiting for the pre-show to start.

Of course, you're not here for that extra news, you're here for Google I/O. From my prior years attending the event, I'll guess that Karissa and Igor are currently gaining sustenance from the press booth and hoping to secure good spots once they're allowed into the seating area. 

Would it be wrong if I shouted out non-Google news here? While we wait? For example, Microsoft is hosting its Ability Summit today, while Apple just announced a bunch of new accessibility features this morning. 

We are 30 minutes out from the "pre-show" and an hour and a half from the start of the keynote and there is already quite a line forming to get into the amphitheater. You can always count on Google IO attendees to be extra hyped.

Oh that picture is the perfect prompt for me to bring up the fact that this Thursday is Global Accessibility Awareness Day. The cap is an AI prompt, sure, but I initially thought it was the alt text for an image. 

I'm not one to normally comment on event swag, but Google gave us these hats, and it's definitely a choice...

I believe Igor has also arrived at Shoreline and will be posting some excellent pictures here soon!

While I sit here at home recapping Google's AI history, Karissa says a marching band has just showed up at Shoreline Amphitheater. Fun! Coachella vibes!

Almost ten years ago, Google CEO Sundar Pichai declared that Google would be shifting its focus "from mobile first to AI-first." It's hard to imagine it's been that long, but when you think about it, what we define as AI today is very different from what it used to be. Many of Google's tools used to be considered AI — things like Smart Replies have existed for years.

As we wait for Karissa and Igor to settle down and get situated at I/O, I'm going to speculate on what we might see today. Obviously AI and new Gemini models are to be expected, and if I were a prediction market person I'd be putting money on Sundar Pichai doing some sort of recap on how far the company's AI products have come. 

Here's Karissa in front of a line of people waiting around the I/O sign at Shoreline Amphitheater. Looks like lovely weather at the moment!

Karissa overheard someone at Shoreline say that I/O is "Coachella for developers." I don't know if I agree, but I think developer season is certainly a vibe.

If Google's keynote does run for about two hours today (like I'm expecting based on past I/Os), that would explain why it had to drop some news last week at the Android Show: I/O edition. The company has so much to share, just one multi-hour event clearly isn't enough.

I hear Karissa has already arrived at I/O and acquired her badge. What a champ!

We are expecting a ton of news from Google across the Cloud, Gemini, Workspace, Search and possibly even its XR divisions today, and my guess is the keynote will last close to two hours. That means you should really be readying a snack and getting a toilet break in before things kick off in just about two and a half hours!

Good morning everyone, and welcome to our liveblog of Google I/O 2026! Our on-the-ground team of senior reporters Karissa Bell and Igor Bonifacic are heading to Shoreline Amphitheater, and I'm sure we'll be seeing pictures and commentary from them here shortly. In the meantime, I wanted to thank you all for joining us this morning. Things look different, but we're still the same people!