Transcript of Google CEO Sundar Pichai at Google I/O 2025 Keynote

Here’s an example of how this could be useful when booking a vacation rental in South America and you don’t speak the language. Let’s take a look.

SPEAKER: Hi, Camilla. Let me turn on speech translation. It’s nice to finally talk to you.

CAMILLA: You’re going to have a lot of fun, and I think you’re going to love visiting the city. The house is in a very nice neighborhood and overlooks the mountain.

SPEAKER: That sounds wonderful.

CAMILLA: There’s a bus nearby, but I would recommend renting a car so you can visit the nature and enjoy it.

SPEAKER: That sounds great.

SUNDAR PICHAI: You can see how well it matches the speaker’s tone, patterns and even their expressions. We are even closer to having a natural and free flowing conversation across languages. And today, we are introducing this real-time speech translation directly in Google Meet. English and Spanish translation is now available for subscribers with more languages rolling out in the next few weeks. And real-time translation will be coming to enterprises later this year.

Project Astra

Another early research project that debuted on the IO stage was Project Astra. It explores the future capabilities of a universal AI assistant that can understand the world around you. We are starting to bring it to our products. Today Gemini Live has Project Astra’s camera and screen sharing capabilities so you can talk about anything you see. People are using it in so many ways, whether practicing for a job interview or training for a marathon.

We’ve been appreciating the feedback from our trusted testers and some who are a little less trusted. Take a look.

USER: That’s a pretty nice convertible.

GEMINI: I think you might have mistaken the garbage truck for a convertible. Is there anything else I can help you with?

USER: What’s this skinny building doing in my neighborhood?

GEMINI: It’s a streetlight, not a building.

USER: Why are these palm trees so short? I’m worried about them.

GEMINI: They’re not short. They’re actually pretty tall.

USER: Sick convertible.

GEMINI: Garbage truck again. Anything else?

USER: Why do people keep delivering packages to my lawn?

GEMINI: It’s not a package. It’s a utility box.

USER: Why is this person following me wherever I walk?

GEMINI: No one’s following you. That’s just your shadow.

SUNDAR PICHAI: Gemini is pretty good at telling you when you’re wrong. We are rolling this out to everyone on Android and iOS starting today.

Project Mariner and Agent Capabilities

Next, we also have our research prototype Project Mariner. It’s an agent that can interact with the web and get stuff done. Stepping back, we think of agents as systems that combine the intelligence of advanced AI models with access to tools. They can take actions on your behalf and under your control.

Computer use is an important agentic capability. It’s what enables agents to interact with and operate browsers and other software. Project Mariner was an early step forward in testing computer use capabilities. We released it as an early research prototype in December, and we have made a lot of progress since.

First, we are introducing multitasking and it can now oversee up to 10 simultaneous tasks. Second, it’s using a feature called teach and repeat. This is where you can show it a task once and it learns a plan for similar tasks in the future. We are bringing Project Mariner’s computer use capabilities to developers via the Gemini API. Trusted testers like Automation Anywhere and UiPath are already starting to build with it, and it will be available more broadly this summer.

Computer use is part of a broader set of tools we will need to build for an agent ecosystem to flourish, like our open agent to agent protocol so that agents can talk to each other. We launched this at Cloud Next with the support of over 60 technology partners and hope to see that number grow. Then there is the model context protocol introduced by Anthropic so agents can access other services. And today, are excited to announce that our Gemini SDK is now compatible with MCP tools. These technologies will work together to make agents even more useful.

Agent Mode in Gemini App

And we are starting to bring agentic capabilities to Chrome, Search, and the Gemini app. Let me show you what we are excited about in the Gemini app. We call it agent mode.

Say you want to find an apartment for you and two roommates in Austin. You’ve each got a budget of $1,200 a month. You want a washer dryer or at least a laundromat nearby. Normally, you’d have to spend a lot of time scrolling through endless listings.

Using Agent Mode, the Gemini app goes to work behind the scenes. It finds listings from sites like Zillow that match your criteria and uses Project Mariner when needed to adjust very specific filters. If there’s an apartment you want to check out, Gemini uses MCP to access the listings and even schedule a tour on your behalf. And it’ll keep browsing for new listings for as long as you need, freeing up to do the stuff you want to do, like plan the housewarming party.

It’s great for companies like Zillow, bringing in new customers and improving conversion rates. An experimental version of the agent mode in the Gemini app will be coming soon to subscribers. This is a new and emerging area, and we are excited to explore how best to bring the benefits of agents to users and the ecosystem more broadly.

Personal Context

The best way we can bring research into reality is to make it really useful in your own reality. That’s where personalization will be really powerful. We are working to bring this to life with something we call personal context. With your permission, Gemini models can use relevant context across your Google apps in a way that is private, transparent, and fully under your control.

Let me show you an example in Gmail. You might be familiar with our AI powered smart reply features. It’s amazing how popular they are. Now imagine if those responses could sound like you. That’s the idea behind personalized smart replies.

Let’s say my friend wrote to me looking for advice. He’s taking a road trip to Utah, and he remembers, I did this trip before. Now if I’m being honest, I would probably reply something short and unhelpful. Sorry, Felix. But with personalized smart replies, I can be a better friend.

That’s because Gemini can do almost all the work for me, looking up my notes in Drive, scanning past emails for reservations, and finding my itinerary in Google Docs, trip to Zion National Park. Gemini matches my typical greetings from past emails, captures my tone, style, and favorite word choices, and then it automatically generates a reply. I love how it included details like keeping driving time under five hours per day, and it uses my favorite adjective, exciting. Looks great. Maybe you want to make a couple of changes to it and hit send. This will be available in Gmail this summer for subscribers. And you can imagine how helpful personal contacts will be across search, docs, Gemini, and more.

Today, I’ve talked about intelligence, agents, and personalization. These are a few of the frontiers where we will make progress, and you’ll hear more examples throughout the keynote. But first, I want to invite someone who can share more about the intelligence driving our future innovation. Last year, I introduced him as Sir Demis. This year, we can add Nobel Laureate to his list of titles. Come on out, Demis.

Demis Hassabis on Gemini Progress

DEMIS HASSABIS: Hey, everyone. It’s really great to be back here at IO. We’re living through a remarkable moment in history where AI is making possible an amazing new future. It’s been a year of relentless progress for us.

As Sundar said, people love interacting with Gemini 2.5. We’ve seen you vibe code with it, building games and apps in a single shot. You’ve used its vast reasoning powers everything from unpacking scientific papers to understanding YouTube videos, and you’ve told us how collaborative, insightful, and genuinely helpful you found using Gemini. It’s so exciting for us to see the awesome things you’ve all been doing and building with it.

Gemini 2.5 Pro is our most intelligent model ever and the best foundation model in the world. Just two weeks ago, we shipped a preview of an updated 2.5 Pro so you could get your hands on it and start building with it right away. We’ve been really impressed by what you’ve created, from turning sketches into interactive apps to simulating entire 3D cities.

The new 2.5 Pro tops the popular coding leaderboard Web Dev Arena. And now that it incorporates LearnLM, our family of models built with educational experts, 2.5 Pro is also the leading model for learning, and it’s number one across all the leaderboards on LM Arena. Gemini Flash is our most efficient workhorse model.

It’s been incredibly popular with developers who love its speed and low cost. Today, I’m thrilled to announce that we’re releasing an updated version of 2.5 Flash. The new Flash is better in nearly every dimension, improving across key benchmarks for reasoning, code, and long context. In fact, it’s second only to 2.5 Pro on the LM Arena leaderboard. I’m excited to say that Flash will be generally available in early June with Pro soon after.

We’re making final adjustments based on your feedback, but you can go try out the preview now in AI Studio, Vertex AI, and Gemini app. As always, we’re looking to push the state of the art relentlessly forward, and we’re excited about our latest research advances. To tell you more, I’d like to welcome Tulsi to the stage, but first, let’s take a look at what you’re building with Gemini today.

“Can you make a bunch of, like, confetti circles explode out of my mouse? I’ve updated the sketch.”

“Okay, what should we build next?”

“Okay, here’s another 30 ideas for you.”

Tulsi: Gemini 2.5 Improvements for Developers

Thanks, Demis. I’m so excited to share the improvements we’re creating to make it easier for developers like all of you to build with Gemini 2.5, all based on your feedback with improved capabilities, enhanced security and transparency, better cost efficiency, and more control.

First, in addition to the new 2.5 Flash that Demis mentioned, we are also introducing new previews for text to speech. These now have a first of its kind multi-speaker support for two voices built on native audio output.

This means the model can converse in more expressive ways. It can capture the really subtle nuances of how we speak. It can even seamlessly switch to a whisper like this. This works in over 24 languages, and it can easily go between languages. So the model can begin speaking in English, but then, and switch back all with the same voice.

That’s pretty awesome, right? You can use this text to speech capability starting today in the Gemini API. The live API will also have a 2.5 Flash preview version of native audio dialogue later today.

So you can directly build conversational experiences with a more natural Gemini. It’s even able to distinguish between the speaker and background voices so it knows when to respond. And Logan and Paige will show you more in the developer keynote.

Second, we’ve strengthened protections against security threats like indirect prompt injections. So Gemini 2.5 is our most secure model yet.

And in both 2.5 Pro and Flash, we’re including thought summaries via the Gemini API and Vertex AI. Thought summaries take the model’s raw thoughts and organize them into a clear format with headers, key details, and information about model actions like tool calls. This gives you increased transparency for what the model is thinking, which is especially useful for high latency tasks while also just being easy to debug and consume.

You also asked us for more efficiency. So today’s 2.5 Flash is even more efficient, reducing the number of tokens the model needs to use for the same performance resulting in 22% efficiency gains on our evals.

Finally, we launched 2.5 Flash with thinking budget to give you control over cost and latency versus quality, and the response was great. So we’re bringing thinking budgets to 2.5 Pro, which will roll out in the coming weeks along with our generally available model. With thinking budgets, you can have more control over how many tokens the model uses to think before it responds, or you can simply turn it off.

Coding with Gemini 2.5 Pro

Now as you heard, Gemini 2.5 Pro is incredible at coding. So now let me show you how you can take any idea you have and bring it to life.

So if you’ve ever been to the American Museum of Natural History in New York City, it has a set of amazing exhibits. To bring that to you today, I got 2.5 Pro to code me a simple web app in Google AI Studio to share some photos and learn more. So here’s what I have. But I want to make it more interactive. And I’m still brainstorming the design, but I’ve got some ideas.

You’ve seen something like this before, right? Someone comes to you with a brilliant idea scratched on a napkin. As a PM, I’m awesome at this.

Now standard two-dimensional web design is one thing, but I wanted to make it three-dimensional. And I learned that jumping into 3D isn’t easy. It requires learning about all kinds of new things, setting up a scene, camera, lighting, and more. Luckily for me, 2.5 Pro can help.

So here, what I’m going to do is I am going to add the image I just showed you with the sphere, and I’m going to add in a prompt that asks 2.5 Pro to update my code based on the image. So we’ll let 2.5 Pro get to work. And as you can see, it’s starting to think, and it’s going ahead and creating a plan based on what I asked for.

And it’ll apply it to my existing code base. Because Gemini is multimodal, it can understand the abstract sphere sketch and code beautiful 3D animations, applying them to my existing app. So this takes about two minutes. So for the purpose of time, we’re going to do this baking show style, and I’m going to jump to another tab that I ran right before this keynote with the same prompt.

And here’s what Gemini generates. Woah. We went from that rough sketch directly to code, updating multiple of my files. And actually, you can see it thought for thirty-seven seconds. And you can see the changes it thought through and then the files it updated. We did all of this in AI Studio.

So once I finished prototyping, I can simply deploy the code along with my Gemini API key. So here’s our final app in Chrome. Look at these animations. And I didn’t need to have advanced knowledge of three.js libraries or figure out the complex 3D math to build this. I know it would have taken forever to do this by hand, and instead, I was able to create this just based on a sketch.

I can make this experience even richer with multimodality. So I use 2.5 Flash to add a question to each photo inviting you to learn a little more about it. But what if it talked? That’s where Gemini’s native audio comes in.

“That’s a pangolin, and its scales are made of keratin just like your fingernails.”

Wow. Now we’re talking. You can hear how you can add expressive audio right into your apps. And before I share more, I’ll leave this demo with another fun layout that 2.5 Pro coded just for us. We put the starter code for anyone to build on top of this demo in Google AI Studio, so go try it out.

And this is just one example of how Gemini 2.5 is changing how we build and how you can leverage the incredible capabilities to integrate vision, speech, and coding into your own applications.

Joules and Gemini Diffusion

2.5 Pro is available on your favorite IDE platforms and in Google products like Android Studio, Firebase Studio, Gemini Code Assist, and our asynchronous coding agent, Joules. Just submit a task and Jules takes care of the rest, fixing bugs, making updates. It integrates with GitHub and works on its own. Jules can tackle complex tasks in large code bases that used to take hours, like updating an older version of Node.js. It can plan the steps, modify files, and more in minutes.

So today, I’m delighted to announce that Joules is now in public beta, so anyone can sign up at joules.google.

And like Demis said, we’re always innovating on new approaches to improve our models, including making them more efficient and performant. We first revolutionized image and video generation by pioneering diffusion techniques.

A diffusion model learns to generate outputs by refining noise step by step. Today, we’re bringing the power of diffusion to text with our newest research model. This helps it excel at tasks like editing, including in the context of math and code. Because it doesn’t just generate left to right, it can iterate on a solution very quickly and error correct during the generation process.

Gemini diffusion is a state-of-the-art experimental text diffusion model that leverages this parallel generation to achieve extremely low latency. For example, the version of Gemini diffusion we’re releasing today generates five times faster than even 2.0 Flashlight, our fastest model so far, while matching its coding performance.

So take this math example. Ready? Go. If you blinked, you missed it. Now earlier, we sped things up, but this time, we’re going to slow it down a little bit. Pretty cool to see the process of how the model gets to the answer of 39.

This model is currently testing with a small group, and we’ll continue our work on different approaches lowering latency in all of our Gemini models with a faster 2.5 Flashlight coming soon. And now for more on the future of Gemini, back to you, Demis.

Demis: DeepThink and World Models

Thanks, Tulsi. We’ve been busy exploring the frontiers of thinking capabilities in Gemini 2.5. As we know from our experience with AlphaGo, responses improve when we give these models more time to think. Today, we’re making 2.5 Pro even better by introducing a new mode we’re calling DeepThink. It pushes model performance to its limits, delivering groundbreaking results.

DeepThink uses our latest cutting-edge research in thinking and reasoning, including parallel techniques. So far, we’ve seen incredible performance. It gets an impressive score on USAMO 2025, currently one of the hardest maths benchmarks. It leads on LiveCodeBench, a difficult benchmark for competition level coding. And since Gemini has been natively multimodal from the start, it’s no surprise that it also excels on the main benchmark measuring this, MU.

Because we are defining the frontier with 2.5 Pro DeepThink, we are taking a little bit of extra time to conduct more frontier safety evaluations and get further input from safety experts. As part of that, we are going to make it available to trusted testers via the Gemini API to get their feedback before making it widely available. You’ll hear more about our plans for 2.5 Pro DeepThink from Josh later today.

Over the past decade, we’ve laid a lot of the foundations for the modern AI era. From our pioneering work inventing the transformers architecture, which today underpins all large language models, to agent systems like AlphaGo and AlphaZero, we continue to double down on the breadth and depth of our fundamental research to invent the next big breakthroughs that are needed for artificial general intelligence.

Gemini is already the best multimodal foundation model, but we’re working hard to extend it to become what we call a world model. That is a model that can make plans and imagine new experiences by simulating aspects of the world just like the brain does. This is something I’ve always been passionate about, having thought about virtual worlds since my early days, developing AI for simulation games like Theme Park.

We’ve been taking strides in this direction for a while through our pioneering work training agents to master complex games like GO and StarCraft, to most recently, our Genie two model capable of generating 3D simulated environments that you can interact with just from a single image prompt.

You can already see these capabilities emerging in the way Gemini can use its world knowledge and reasoning to represent things in nature. And in VEO, our state-of-the-art video model, which has a deep understanding of intuitive physics, like how gravity, light, and materials behave. It’s really incredible how well VIO maintains accuracy and consistency across frames. It knows what to do even when the prompts get a little creative, like this person made out of life rafts.

Understanding the physical environment will also be critical for robotics. AI systems will need world models to operate effectively in the real world. We fine-tuned a specialized model, Gemini Robotics, that teaches robots to do useful things like grasp, follow instructions, and adjust to novel tasks on the fly. For those of you here today, you can play around with the robots in the AI sandbox.

Project Astra and the Future of Gemini

Making Gemini a full world model is a critical step in unlocking a new kind of AI, one that’s helpful in your everyday life, that’s intelligent and understands the context you’re in, and that can plan and take action on your behalf across any device. This is our ultimate vision for the Gemini app, to transform it into a universal AI assistant, an AI that’s personal, proactive, and powerful, and one of our key milestones on the road to AGI.

This starts with the capabilities we first explored last year in Project Astra, such as video understanding, screen sharing, and memory. Over the past year, we’ve been integrating capabilities like these into Gemini Live for more people to experience today.

Of course, we continue to relentlessly improve and explore new innovations at the frontier. For example, we’ve upgraded voice output to be more natural with native audio. We’ve improved memory and added computer control. Let’s take a look.

“I’m here with the latest research prototype of Project Astra, and we’re going see how it can help me fix my bike.”

“Hey. Go online and look for a user’s manual for a Huffy mountain bike.”

“I have found a user manual for you. What would you like to know about it?”

“Okay, cool. Now scroll until you find the section about the brakes.”

“Sure thing. Looking for that page now.”

“I think I stripped this screw. Can you go on YouTube and find a video for how to fix that?”

“Of course. I’m opening YouTube now. This looks like a good video. Place a rubber band over the head of the screw and press down.”

“Can you go through my emails with the bike shop and find which size hex nut I need?”

“Sure. According to the email from Bicycle Habitat, you need a three-eighths inch hex nut for your sprocket. I’ve highlighted the correct bin for you.”

“It seems like I need a spare tension screw. Can you call the nearest bike shop and see what they have in stock?”

“Yep. Calling them now. I’ll get back to you with what they have in stock.”

“I think my brake pads are wearing down. Can you check the manual again and see if it talks about them anywhere?”

“You’ll see how to replace the brake pads.”

“Hey. Any updates on that call?”

“Yep. I just got off of the bike shop.”

AI Mode in Google Search: Transforming How We Find Information

SUNDAR PICHAI: A universal AI assistant will perform everyday tasks for us. It will take care of mundane admin and surface delightful new recommendations, making us more productive and enriching our lives.

We’re gathering feedback about these new capabilities now from trusted testers and working to bring them to Gemini Live, new experiences in search, and the live API for developers, as well as new form factors like Android XR Glasses. You’ll hear more on this later today.

My entire career at its core has been about using AI to advance knowledge and accelerate scientific discovery. At Google DeepMind, we’ve been applying AI across almost every branch of science for a long time. In just the past year, we’ve made some huge breakthroughs in a wide range of areas from mathematics to life sciences.

We’ve built AlphaProof that can solve mass Olympiad problems at the silver medal level. Coscientists that can collaborate with researchers helping them develop and test novel hypotheses. And we’ve just released AlphaEvolve, which can discover new scientific knowledge and speed up AI training itself.

In the life sciences, we’ve built AIMI, a research system that could help clinicians with medical diagnoses, AlphaFold three, which can predict the structure and interactions of all of life’s molecules, and isomorphic labs, which builds on our AlphaFold work to revolutionize the drug discovery process with AI and will one day help to solve many global diseases.

In just a few short years, AlphaFold has already had a massive impact in the scientific community. It’s become a standard tool for biology and medical research, with over 2.5 million researchers worldwide using it in their critical work.

As we continue to make progress towards AGI, I’ve always believed, if done safely and responsibly, it has the potential to accelerate scientific discovery and be the most beneficial technology ever invented.

Taking a step back, it’s amazing to me that even just a few years ago, the frontier technology you’re seeing today would have seen nothing short of magic. It’s exciting to see these technologies powering new experiences in products like Search and Gemini and also coming together to help people in their daily lives.

For example, we recently partnered with Aira, a company that assists people in the blind and low vision community to navigate the world by connecting them via video to human visual interpreters. Using Astra technology, we built a prototype to help more people have access to this type of assistance. We’re getting ongoing feedback from users while Iris interpreters are actively supervising for safety and reliability.

With this and all our groundbreaking work, we’re building AI that’s more personal, proactive, and powerful, enriching our lives, advancing the pace of scientific progress, and ushering in a new golden age of discovery and wonder.

Project Astra: Enhancing Accessibility

I remember as a child playing till it got dark outside to make sure I could still play guitar without being able to see. I was diagnosed with retinitis pigmentosa when I was probably about four years old. My vision is actively deteriorating, but you know music has been something that I can continue to do with the closing in of my visual reality. I feel free when I play music.

If Project can help me be more independent, it would be, you know, the greatest thing ever. Touring, for example, I love it because I get to go out and go to places I’ve never been before and meet new people, but it definitely is becoming more difficult.

ASTRA: Hello there.

USER: Hello. Can you tell me what you see in this green room while I scan around?

ASTRA: I see a sign on the wall in the direction you’re facing. The sign says Wi-Fi network, the Grey Eagle, and the password is live music.

USER: Thank you very much. Can you help me locate a microphone stand?

ASTRA: Sure, I can help with that. Do you mind if I turn on the flashlight?

USER: Yes, please.

ASTRA: I see a wall with several coiled cables and what looked like Mike stands towards the top right in the video.

USER: The most powerful thing I can do is to get on stage, pick up my guitar and play. It helps people understand that there’s more than just blind or not blind disability and ability. If project Astra could help me along the way, I’m all for it.

ASTRA: Have a great time at your show.

USER: Thanks, Dennis.

SUNDAR PICHAI: Amazing to see the possibilities for Project Astra to help with accessibility. It’s another exciting example of how AI is advancing our timeless mission to organize the world’s information and make it universally accessible and useful.

No product embodies our mission more than Google Search. It’s the reason we started investing in AI decades ago and how we can deliver its benefits at the scale of human curiosity. Our Gemini models are helping to make Google Search more intelligent, agentic, and personalized.

One great example of progress is our AI overviews. Since launching at IO last year, they have scaled up to over 1.5 billion users every month in more than 200 countries and territories. As people use AI Overviews, we see they are happier with their results, and they search more often. In our biggest markets like The US and India, AI Overviews are driving over 10% growth in the types of queries that show them. What’s particularly exciting is that this growth increases over time. It’s one of the most successful launches in search in the past decade.

AI overviews are also one of the strongest drivers of growth for visual searches in Google Lens. Lens grew 65% year over year with more than 100 billion visual searches already this year. So people are asking more queries, and they’re also asking more complex queries.

With our latest Gemini models, our AI overviews are at the quality and accuracy you’ve come to expect from Search and are the fastest in the industry. For those who want an end to end AI search experience, we are introducing an all new AI mode. It’s a total reimagining of search.

With more advanced reasoning, you can ask AI mode longer and more complex queries like this. In fact, users have been asking much longer queries, two to three times the length of traditional searches. And you can go further with follow-up questions. All of this is available today as a new tab right in search. I’ve been using it a lot, and it’s completely changed how I use search. And I’m excited to share that AI mode is coming to everyone in The US starting today.

AI mode is where we will first bring our frontier capabilities into search. And starting this week, Gemini 2.5, our most intelligent model series, is coming to search. To share a whole lot more, here’s Liz.

The Future of Google Search

LIZ: Thanks, Sundar. In just one year, there’s been a profound shift in how people are using search. For years, people have come to Google to ask questions like this, and now they’re also asking questions that look more like this. They’re asking longer questions, harder ones, and as you just heard from Sundar, lots more of them.

They’re experiencing what AI powered search can do as we bring together our Gemini model’s advanced capabilities with search’s unparalleled understanding of the web and the world’s information. Today, you’ll see how you can ask anything, and a more intelligent, agentic, and personalized search will take on your toughest questions and help you get stuff done.

This is the future of Google Search, a search that goes beyond information to intelligence. And you’re starting to see this come to life already with AI overviews, and AI mode takes this to the next level. AI mode is search transformed with Gemini 2.5 at its core. It’s our most powerful AI search, able to tackle any question. And as Sundar announced, we’re excited to start rolling out AI mode for everyone in The US starting today.

You’ll find it as a new tab directly in search or right from your search bar. AI mode will be loaded up with all of our best AI features and capabilities, but it’s even more than that. It’s a glimpse of what’s to come. Over time, we’ll graduate many of AI Mode’s cutting edge features and capabilities directly into the core search experience.

That starts today as we bring the same models that power AI Mode to power AI overviews so you can bring your hardest questions right to the search box. Today, we’ll give you a tour of AI mode, and you’ll see how it works and how it’s getting even better with personal context, deeper research, complex analysis and visualization, live multimodality, and new ways to shop.

Now that’s a lot because AI mode can do a lot. So let’s dive in. First, with AI mode, you can ask whatever is on your mind. And as you can see here, search gets to work. It generates your response, putting everything together for you, including links to content and creators you might not have otherwise discovered and merchants and businesses with useful information like ratings.

Search uses AI to dynamically adapt the entire UI, the combination of text, images, links, even this map just for your question. And you can follow-up conversationally. Now AI mode isn’t just giving you information. It’s bringing a whole new level of intelligence to search.

What makes this possible is something we call our query fan out technique. Now under the hood, search recognizes when a question needs advanced reasoning. It calls on our custom version of Gemini to break the question into different subtopics, and it issues a multitude of queries simultaneously on your behalf.

It searches across the entire web going way deeper than a traditional search. And it taps into all of our datasets of real time information like the knowledge graph, the shopping graph, and in this case, local data, including insights from our maps community of over 500 million contributors. Search pulls together a response and it checks its work to make sure it meets our high bar for information quality. And if it detects any gaps, it issues even more searches to fill them in. That means with AI mode, you get all of this from just a single search, and you get it fast.

Personalized AI Mode

Now let’s take a look at what’s coming next to AI Mode starting in Labs. Soon, AI Mode will be able to make your responses even more helpful with personalized suggestions based on your past searches. You can also opt in to connect other Google apps starting with Gmail. We call this personal context and you’ll see when AI mode is bringing yours in to help.

So now based on your recent restaurant bookings and searches, it gets that you prefer outdoor seating. And since you subscribe to those gallery newsletters, it suggests some cool art exhibits to check out while you’re in town. But that’s not all. Because your flight and hotel confirmations are in your inbox, you get event ideas to sync up with when you’ll actually be in Nashville with many nearby where you’re staying.

You can see how personal context in AI mode makes search really yours with recommendations customized just for you. Now this is always under your control and you can choose to connect or disconnect at any time. Personal context is coming to AI mode this summer.

Deep Search Capabilities

Next, for questions when you want an even more thorough response, we’re bringing deep research capabilities into AI mode. You already come to search today to really unpack a topic, but this brings it to a much deeper level, so much so that we’re calling this deep search.

Deep search uses the same query fan out techniques you just heard about but multiplied. It can issue dozens or even hundreds of searches on your behalf. It reasons across all those disparate pieces of information to create an expert level, fully cited report in just minutes. It includes links to the web throughout so you can easily explore and take action.

Now that’s a core part of how we’ve built AI mode overall and how we’ve always thought about AI in search because we believe AI will be the most powerful engine for discovery that the web has ever seen, helping people discover even more of what the web has to offer and find incredible hyper relevant content. You’re starting to see how search is becoming more intelligent, and we’ve got more to show you. So I’ll hand it over to Raja, who will share how AI Mode is helping analyze complex data for one of his passions, sports.

Complex Analysis and Data Visualization

RAJA: Thanks, Liz. Okay. So I’m a huge baseball fan, and lately, there’s been a lot of buzz about these new torpedo bats. If you don’t follow baseball, it’s a new bat design where more of the weight of the bat is in the sweet spot.

As you can see, I’ve been digging in on whether it’s making a real impact on the game, and now I’m wondering what the numbers say. So I’ll ask, show the batting average and on base percentage for this season and last for notable players who currently use a torpedo bat.

Think about it. There are so many parts to that question. Search needs to understand who the notable players are, which ones are using torpedo bats, and their stats. I get this helpful response including this easy to read table. And I know that this is fresh and accurate since it uses our sports data that’s continuously updated down to the last strike.

Search even brings in important context, like that it’s still early in the season. I can follow-up and ask, how many home runs have these players hit this season? And just like that, I get this graph. This goes back to what Liz mentioned about AI mode dynamically generating the right UI for each response.

Search figured out that the best way to present this information is a graph, and it created it. It’s like having my very own sports analyst right in search. Complex analysis and data visualization is coming this summer for sports and financial questions.

Okay. So all this talk about baseball made me want to get closer to the game, like at the next game close. But finding the perfect tickets can be a chore. So I’m excited to share that we’re bringing Project Mariner’s agentic capabilities into AI mode. You’ve already seen how AI mode is becoming more intelligent and personalized.

# Google I/O ’25 Keynote Transcript (Continued)

Search Gets More Agentic

And here’s where you start to see search getting more agentic. Search can take work off my plate while still under my control. I’ll say, find two affordable tickets for this Saturday’s Reds game in the lower level. Search kicks off a query fan out, looking across several sites to analyze hundreds of potential ticket options. Doing the tedious work of filling in forms with all the criteria I asked for.

And it puts it all together, reasoning across the results to analyze real-time pricing and inventory. Then right here, task complete. I get great ticket options with helpful context so I can make an informed decision. Looks like these seats have a good view and at a reasonable price. Search helps me skip a bunch of steps, linking me right to finish checking out.

Tickets secured. Search will help with tasks like this soon, starting with event tickets, restaurant reservations, and appointments for local services.

Multimodal Search Evolution

Next, let’s talk about multimodality. We’ve been blazing the trail for multimodal search since before it was really even a thing. We introduced Google Lens on this very stage back in 2017, and since then, we’ve made it even easier to search what you see.

Snap a picture with Google Lens or simply circle the search, and you can get an AI overview instantly. Like Sundar mentioned, visual search is on fire, and today, I’m excited to share that Lens has over 1.5 billion users every month. Now we’re taking the next big leap in multimodality by bringing Project Astra’s live capabilities into AI mode.

Think about all those questions that are so much simpler to just talk through and actually show what you mean, like a DIY home repair, a tricky school assignment, or learning a new skill. We call this Search Live, and now using your camera, search can see what you see and give you helpful information as you go back and forth in real time.

It’s like hopping on a video call with search. I have three kids and they ask about a million questions a minute. And with summer right around the corner, the team and I decided to put Search Live to the ultimate test, helping us and our kids tackle something new. We recorded at home with our families just this past weekend. Let’s take a look.

PRESENTER: It looks like you’re about to do a fun science experiment.

PARENT: Alright. Ready Anya? Ready, Adi?

CHILDREN: Yeah.

PARENT: Okay. Are you ready for your science experiment?

CHILD: Yeah.

PARENT: Can you guess which experiment I’m trying to do? I have hydrogen peroxide and dish soap and yeast.

PRESENTER: You’re likely going to make elephant toothpaste.

PARENT: How do I know which one of these strawberries is regular?

PARENT: I’m trying to get this to your board tour.

PRESENTER: It looks like someone is ready to get their hands dirty and plant a green bean seedling up to about 50 pounds per square inch, but don’t go over 90 PSI.

PARENT: I mixed baking soda. What do I do next?

PRESENTER: It looks like someone tried dipping a finger into the cinnamon water. What should I do to make this even more impressive?

PRESENTER: You could try using a different catalyst. Potassium iodide is a good option.

CHILD: Woah. Why is it doing that?

PRESENTER: A ripe strawberry will also have a sweet smell. The green leaves at the top should look fresh and vibrant. The chemical reaction is going well.

PARENT: This is awesome. Thanks.

CHILD: Woah. Can we do it again?

PARENT: Go.

PARENT: Oh, boy. How do I get strawberry stains out of a shirt?

PRESENTER: Try using a mix of one tablespoon white vinegar, half a teaspoon liquid laundry detergent.

AI-Powered Shopping

With AI mode, we are bringing a new level of intelligence to help you shop with Google. As you heard earlier, AI Mode brings in information from the web and our real-time data.

In this case, that means you get the visual inspiration of Google Images and the world’s most comprehensive set of products and retailers from our shopping graph, which has over 50 billion product listings that get constantly updated. Let me show you how this comes together.

So I’ve been on the hunt for a new rug for my living room. I’ll say, I have a light gray couch, and I’m looking for a rug to brighten the room. What I need first is visual inspiration.

Search dynamically generates a browsable mosaic of images and some shoppable products personalized just for me. I see rugs from some of my go-to brands and some more modern options since I often search for things in that style. Now these are nice, but I know most of them wouldn’t survive a day in my home. So I’ll add, I have four active kids, and they like to have friends over.

Search understands this involves plenty of messes. It recommends a low pile and washable rug, made from durable materials along with some products that I can shop. But I don’t know a lot about rugs, so luckily, Search points out some things to consider and with helpful follow-ups like these, taking a few days of research down to just a few minutes.

You can imagine how helpful this new experience in AI mode will be for all of your shopping needs. Speaking of, I want a few new dresses for summer. Search gave me some great options, but now I’m faced with the classic online shopping dilemma.

Virtual Try-On Feature

I have no clue how these styles will look on me. So we are introducing a new try-on feature that will help you virtually try on clothes so you get a feel for how styles might look on you. Let me show you a live demo.

Okay. So I look at many dresses, and I really like this blue one. I click on this button to try it on. It asks me to upload a picture, which takes me to my camera roll. I have many pictures here. I’m going to pick one that is full length and a clear view of me, and off it goes.

While it’s processing, I’ll show you what’s happening behind the scenes. To create a try-on experience that works at scale, we need a deep understanding of the human body and how clothing looks on it. To do this, we built a custom in a generation model specifically trained for fashion.

Wow. And it’s back. I have to say I love a live demo when it works.

Alright. More on how it works. It brings in advanced 3D shape understanding, which allows us to perceive shapes and depths more accurately, helping us better visualize the human body. Our try-on experience works with your photo. It’s not some pre-captured image or a model that doesn’t look like you.

And then when it comes to clothes that you’re interested in, the AI model is able to show how this material will fold and stretch and drape on people. This technology is the most state-of-the-art in the industry at scale, and it allows us to visualize how billions of apparel products look on a wide variety of people. And you can see it here how it really gives me a feel for how this dress might look on me.

Agentic Checkout Feature

Alright. So I’m now set on the dress, and Search can help me find it at the price that I want and buy it for me with our new agentic checkout feature.

So let me get back here to the dress, and I’m going to click this thing to track price. I pick my size. Then after set a target price, I’m going to set it to about $50. And tracking is happening. Search will now continuously check websites where the address is available and then let me know if the price drops.

So now let’s switch out of our live demo mode, and then I’m gonna sprinkle some IO magic. And let’s assume the price has now dropped. When that happens, I get a notification just like this. And if I want to buy, my checkout agent will add the right size and color to my cart. I can choose to review all my payment and shipping information or just let the agent just buy it for me.

With just one tap, search, securely buy this for me with Google Pay. And, of course, all of this happened under my guidance from inspiration to purchase. Our new intelligent capabilities bring the best of shopping together with the best of AI right here in Search. Our new visual shopping and agency checkout features are rolling out in the coming months, and you can start trying on Looked in labs beginning today. And now back to you, Liz.

LIZ: Thanks, Vidya. As Sundar mentioned, this all gets an even bigger upgrade later this week when we bring a custom version of Gemini 2.5 to both AI overviews and AI mode. This is the next frontier of Google Search where AI overviews bring the helpfulness of AI to everyone in the main search experience. And as you’ve seen today, AI mode is a totally reimagined AI search with all our most cutting-edge AI features and capabilities.

We couldn’t be more excited about this chapter of Google Search where you can truly ask anything, and we mean anything. Your simplest and your hardest questions, your deepest research, your personalized shopping needs, your just take it off my plate tasks.

PRESENTER: What else should I do to make it stronger?

PRESENTER: To make it stronger, consider adding more triangles to the design.

Gemini: The Personal AI Assistant

For years, people have pursued building an AI assistant that doesn’t just respond but understands, one that doesn’t just wait but anticipate, a truly universal assistant that empowers you. And today, we’re taking a new step forward to that future, and we’re using the Gemini app to get there.

Our goal is to make Gemini the most personal, proactive, and powerful AI assistant, and it starts with being personal. What if your AI assistant was truly yours, an assistant that learns you, your preferences, your projects, your world, and you were always in the driver’s seat. And with your permission, you could share relevant Google information with it, making it an extension of you.

We call this personal context, and we’ve already made it possible for you to connect your search history with Gemini so it can understand that you’ve searched for recipes in the past and craft responses like this. Starting soon, you’ll be able to add even more personal context from across Google to Gemini so it will be uniquely helpful.

You can choose to turn this on, and you’ll always be in control of the experience. You can view and manage your information as well as connect and disconnect different Google apps. This level of personalization gives you a more proactive AI assistant, and this changes a lot.

See, today, most AI is reactive. You ask, it answers. But what if it could see what’s coming and help you prepare even before you ask? Imagine you’re a student. You’ve got a big physics exam looming, and instead of scrambling, Gemini sees it on your calendar a week out. But it doesn’t just remind you. It comes with personalized quizzes crafted from your materials, notes from your professor, even photos, handwritten notes.

That’s not just helpful. It’s going to feel like magic. And imagine the step beyond that where Gemini can go off and make custom explainer videos based on your interests so you can understand a topic better. In this case, Gemini knows that I’m into cycling, so it explains this concept of thermodynamics to me using an analogy I’ll understand.

This is where we’re headed with Gemini, and it’s all possible because of the powerful capabilities in the underlying model.

Last month, we shipped our latest 2.5 Pro model in the Gemini app, and it wasn’t just an upgrade. It was a leap forward. People are doing extraordinary things in the app. They’re not just vibe coding. They’re inventing entire games and making fully featured websites in minutes.

They’re thinking things into existence. And it’s fun to write software for the first time or create a video with our VO model to generate and edit images. It’s a new superpower for everyone. And so these three Ps, personal, proactive, and powerful, these are the frontiers of AI assistance. And thanks to 2.5 Pro, we’re making big strides across all three.

New Gemini Features

So let’s talk more about how all this is coming together to life in the Gemini app. We’re launching five things today.

First, let’s talk about Gemini Live. People are blown away by how interactive and natural the conversations are, and it works in over 45 languages, more than 50 countries. It’s so intuitive, so engaging. The conversations, in fact, are five times longer than the text conversations in the app. And I can tell you from personal experience, it’s great for talking through things on the drive in to work in the morning.

Now as Sundar mentioned, Gemini Live now includes camera and screen sharing, both of which are incredible. All of it is rolling out free of charge in the Gemini app on Android and iOS today. And in the coming weeks, you’ll be able to connect Gemini Live to some of your favorite apps, like Calendar, Maps, Keep, Tasks.

So soon, you can just point your camera and ask it to add an invite to your calendar, and it’ll be done. Or if you need to decipher your roommate’s handwriting for the shopping list, Gemini Live can turn those scribbles into a neat list in Google Keep.

Our Gemini RealLife roadmap is overflowing with exciting things. They’re all being prototyped in Project Astra like you saw earlier. And as those ideas mature, we’ll graduate them into Gemini Live for everyone. And since Gemini and Android work so closely together, many of those experiences will work great on Android across the entire ecosystem. So stay tuned for more.

Alright. Real-time interaction is amazing, but sometimes you need to go deep, unravel something complex. This is where deep research comes in.

Starting today, deep research will now let you upload your own files to guide the research agent, which is one of the top requested features. And soon, we’ll let you research across Google Drive and Gmail so you can easily pull in information from there too.

So let’s say you have this incredible detailed report. In this case, it’s about the science of comets moving throughout space. How do you get all that brilliance distilled down into something digestible, engaging, something you can share?

# Gemini’s Canvas: Interactive Co-Creation Space

SUNDAR PICHAI: This is where Canvas comes in. It’s Gemini’s interactive space for co-creation. Canvas will now let you transform that report with one tap into all kinds of new things, like a dynamic web page, an infographic, a helpful quiz, even a custom podcast in 45 languages. But if you want to go further, you can vibe code all sorts of amazing things in Canvas with as much back and forth as you want. You can get exactly the experience you’re looking for.

Check out this interactive comet simulation that one of our Googlers made just by describing what they wanted to build and collaborating with Gemini to get it just right. And as you can share apps like this, others can easily jump in and view it and modify it and remix it. This is the power to transform anything, and it’s a whole new way to use Gemini.

There’s another new way you can use Gemini today too. We’re introducing Gemini in Chrome. This will be your AI assistant that’s there for you as you browse the web on your desktop. The amazing part is that you can use this and it understands the context of the page that you’re on automatically. So if you have a question, it can be answered. I especially love it for comparing reviews on long pages like this camping website. We’re starting to roll out Gemini and Chrome this week to Gemini subscribers in the U.S.

All right. We’ve taken some quizzes. We’ve talked to Gemini Live. We’ve learned about comets and campsites. Now it’s time to create some things.

Imagine 4: Advanced Image Generation

Starting today, we’re bringing our latest and most capable image generation model into the Gemini app. It’s called Imagine 4, and it’s a big leap forward. The images are richer with more nuanced colors and fine-grained details. The shadows in the different shots, the water droplets that come through in the photos.

I’ve spent a lot of time around these models, and I can say this model and the progression has gone from good to great to stunning. And Imagine 4 is so much better at text and topography. In the past, you might have created something that looks good, but adding words didn’t always work just right.

So check this out. Maybe I want to create a poster for a music festival. We’ll make the Chrome dino the big headliner. Imagine 4 doesn’t just get the text and spelling right, it’s actually making creative choices like using dinosaur bones in the font or figuring out the spacing, the font size, the layout that makes it look like this great poster.

So the image quality is higher. The speed is faster. The text is better. All of this lets you make posters, party advice and anything else. And with Gemini’s native image generation, you can easily edit these images too right in the app.

We’ve also made a super fast variant of Imagine 4. We can’t wait for you to get your hands on it. In fact, it’s 10 times faster than our previous model, so you can iterate through many ideas quickly.

VO3: Next-Generation Video with Audio

Alright. I want to show you one thing. Images are incredible, but sometimes you need motion and sound to tell the whole story. Last December, VO2 came out, and it redefined video generation for industry. And if you saw Dennis’ Sizzling Onions post yesterday, you know that we’ve been cooking something else.

Today, I’m excited to announce our new state-of-the-art model, VO3. And like a lot of other things you’ve heard about from stage today, it’s available today. The visual quality is even better. Its understanding of physics is stronger. But here’s the leap forward: VO3 comes with native audio generation. VO3 can generate sound effects, background sounds and dialogue. Now you prompt it and your characters can speak.

Here’s a wise old owl and a nervous young badger in the forest. Take a listen.

“They left behind a ball today. It bounced higher than I can jump.”

“What manner of magic is that?”

Pretty cool, right? VO3 added not just the sounds of the forest, but also dialogue. We’re entering a new era of creation with combined audio and video generation that’s incredibly realistic. The quality is so good. It feels like you’re there on the boat with this guy.

“This ocean, it’s a force, a wild, untamed might, and she commands your awe with every breaking light.”

The photorealistic generation, the emotion, the movement of his mouth and the ocean in the background, it’s incredible how fast VO3 continues to evolve as a powerful creative tool. And we’ve been working closely with the film industry to imagine what’s possible with VO3 and to get this right, and you’ll hear more about that in a few minutes.

Gemini Updates Summary

So that’s what’s happening around Gemini. Starting today, Gemini live capabilities are free and rolling out across Android and iOS. Deep Research and Canvas are getting their biggest updates yet. There’s a new agent mode coming for multistep actions that you heard about earlier. We’re introducing Gemini in Chrome to help you navigate the web, and you can create stunning images and videos with sound using the new Imagine 4 and VO3 models. It’s all coming together in the Gemini app as we work to deliver the most personal, proactive, and powerful AI assistant.

And now to go into more detail about how these generative models are unlocking creative expression, here’s Jason.

# Generative Media for Creative Expression

JASON: Thanks, Josh.

Whether you’re a creator, a musician, or a filmmaker, generative media is expanding the boundaries of creativity. By working closely with the artistic community since the very beginning, we continue to build technology that empowers their creative process.

For example, we worked with musicians to develop Music AI Sandbox. It’s a tool for professionals to explore the possibilities of our generative music model, Lyria, in their work. Let’s watch a clip starring the legendary Grammy award winning singer and composer Shankar Mahadevan putting Music AI Sandbox and Lyria to use.

SHANKAR MAHADEVAN: I’m Shankar Mahadevan, and I’m a music composer, singer, producer, and a happy man. Working at Sandbox is great. We inputted our requirement, and it gave us a bed. We used that bed, and we came up with this song. For a musician, it’s such an inspiring tool. You open a door and you see, hey. There’s another room there. And then you open one more door and you say, more room there. So that’s what AI does. Amazing.

JASON: We recently launched Lyria 2, which can generate high fidelity music and professional grade audio. The music is melodious with vocals in solos and choirs. As you hear, it makes expressive and rich music. Lyria 2 is available today for enterprises, YouTube creators, and musicians.

AI Content Watermarking with SynthID

This focus on collaboration carries over not just into what we build, but how. Right now, it’s not easy for people or organizations to detect AI generated images. This will become only more true as technology improves and the music, audio, images, and videos become more lifelike. So we’re continuing to innovate in this space.

Two years ago, we pioneered SynthID, which embeds invisible watermarks into generated media. To date, over 10 billion pieces of content have been watermarked. We’re also expanding our partnerships to ensure that more content is watermarked with SynthID and that more organizations can detect it too.

Finally, we’re also making it easier to detect the watermark. Our new SynthID detector can identify if an image, audio track, text, or video has SynthID in it, whether it’s in the whole piece or even just a part. We’re starting to roll this out to early testers today.

Collaboration with Filmmakers

Our collaboration efforts have also helped us explore VO3 as a filmmaking tool. Recently, we teamed up with visionary director Darren Aronofsky and his new storytelling venture, Primordial Soup, to shape VO3’s capabilities to meet the needs of storytellers, putting artists in the driver’s seat of innovation.

Together, we are putting the world’s best video model into the hands of top filmmakers to push the boundaries of technology as a tool for more creative and emotional storytelling. The first of the partnership’s three short films is director Eliza McNitt’s “Sphestra.” Let’s take a peek.

ELIZA MCNITT: Film has always been this deeply human act of connecting people with each other’s stories. It has the ability to rip us out of our experience and take us on another journey. I don’t think that ever changes.

This is a story that traverses the entire history of the universe. But it’s really about a mom and what happens when her child is born with a hole in her heart. We filmed really emotional performances, but then generated video we could never capture otherwise.

I want the baby to be holding the mother’s finger. Just the bliss of the two of them.

VO3 is a generative video model, but to me, it’s another lens through which I get to imagine the universe around models, the limits of the models and try to shape them to make storytelling tools.

To be honest, I never thought about telling the story of the day I was born, but here we are. This was supposed to be a checkup. Crash C-section. Baby’s in distress. We need help here. We need to go.

“Is my baby going to be okay?”

“We’re going to do everything we can.”

“Am I going to be okay?”

JASON: Incredible work. Eliza combined live action performance with generated video, like the microscopic worlds, the cosmic events, and herself as a newborn. This approach opened up completely new storytelling possibilities, empowering Eliza to bring cinematic scale and emotion to a deeply personal story.

This partnership also helped shape VO3’s capabilities to meet the needs of storytellers like Eliza. We built new capabilities for filmmakers. So when you’re making a video, it will use ingredients you give it, characters, scenes, or styles, and keep them consistent. Or you can direct video, giving it precise camera instructions and have it shoot along a specific path. These capabilities help filmmakers fluidly express their ideas with VO3.

We then took things one step further. To tell you more, let’s get Josh back out here.

Introducing Flow: AI Filmmaking Tool

JOSH: Thanks, Jason. Based on our collaboration with the creative community, we’ve been building a new AI filmmaking tool for creatives, one that combines the best of VO3, Imagine, and Gemini, a tool built for creatives by creatives. It’s inspired by that magical feeling you get when you get lost in the creative zone and time slows down. We’re calling it Flow, and it’s launching today.

Let me show you how it works. Let’s drop into a project I’m working on. Our hero, the grandpa, is building a flying car with help from a feathered friend. These are my ingredients, the old man and his car. We make it easy to upload your own images into the tool, or you can generate them on the fly using Imagine, which is built right in.

We can create a custom gold gear shift just by describing it. There it is. Pretty cool. Next, you can start to assemble all of those clips together. With a single prompt, you can describe what you want, including very precise camera controls.

Flow puts everything in place, and I can keep iterating in the scene builder. Now here’s where it gets really exciting. If I want to capture the next shot of the scene, I can just hit the plus icon to create the next shot. I can describe what I want to happen next, like adding a 10-foot tall chicken in the back seat, and Flow will do the rest. The character consistency, the scene consistency, it just works.

And if something isn’t quite right, no problem. You can just go back in like any other video tool and trim it up if it’s not working for you. But Flow works in the other direction as well. It lets you extend a clip too, so I can get the perfect ending that I’ve been working towards.

Once I’ve got all the clips I need, I can download the files. I can bring them into my favorite editing software, add some music from Lyria, and now the old man finally has his flying car. Pretty awesome.

I want you to hear just a few words from a few of our AI filmmakers who we’ve been working with to shape Flow, how they describe how it feels using it, and how they’re making it even better. Let’s take a watch.

FILMMAKER 1: Trying anymore, I’m just doing. And all the pieces start falling into place.

FILMMAKER 2: It all feels pretty clear in my head. I see these like flashes of possibilities, almost like I’m traveling through dimensions.

FILMMAKER 3: I’m looking down at myself and my characters in these different worlds and it’s almost coming to life on their own even though I know I’m in control of that narrative.

FILMMAKER 4: It feels like it’s almost building upon itself at some point. You know, you could have an infinite amount of endings to your story.

FILMMAKER 5: So the work isn’t built break by break. It blooms like a spontaneous garden. It grows naturally, fully vibrant and complete.

FILMMAKER 6: I’m not forcing it. I’m just finding it. And that’s when I know I’m in the right place.

New Google AI Subscription Plans

JOSH: So amazing to hear those filmmakers talk about bringing their incredible visions to life. And that feeling of building, of creating, that’s exactly what we want to put in your hands today.

So I’m excited to share that we’re upgrading two AI subscription plans today. We will have Google AI Pro and the all new Google AI Ultra. With the Pro plan, which is going to be available globally, you’ll get a full suite of AI products with higher rate limits and special features compared to the free version. This includes the pro version of the Gemini app that was formerly known as Gemini Advanced.

Then there’s the Ultra plan. It’s for the trailblazers, the pioneers, those of you who want cutting edge AI from Google. The plan comes with the highest rate limits, the earliest access to new features and products from across Google. It’s available in the US today, and we’ll be rolling it out globally soon. You can think of this Ultra plan as your VIP pass for Google AI.

So if you’re an Ultra subscriber, you’ll get huge rate limits and access to that 2.5 pro deep think mode in the Gemini app when it’s ready.

# Google I/O ’25 Keynote (Continued)

Google AI Plans and Subscription Tiers

You’ll get access to Flow with v3 available today, and it also comes with YouTube Premium and a massive amount of storage. We can’t wait to see what you build, create and discover with these new Google AI plans.

Android and XR: The Future of AI in the Physical World

SHARAM: Hi, everyone. There’s so many exciting things happening in Android right now.

It’s the platform where you see the future first. Just last week at the Android show, we unveiled a bold new design and major updates to Android 16 and Wear OS 6. And, of course, Android is the best place to experience AI. Many of the Gemini breakthroughs you saw today are coming soon to Android. You can already access Gemini instantly from the power button.

It understands your context and is ready to help. But Android is powering more than your phone. It’s an entire ecosystem of devices. In the coming months, we’re bringing Gemini to your watch, your car’s dashboard, even your TV. So wherever you are, you have a helpful AI assistant to make your life easier.

But what about emerging form factors that could let you experience an AI assistant in new ways? That’s exactly why we’re building Android XR. It’s the first Android platform built in the Gemini era, and it supports a broad spectrum of devices for different use cases, from headsets to glasses and everything in between. We believe there’s not a one size fits all for XR, and you’ll use different devices throughout your day. For example, for watching movies, playing games, or getting work done, you’ll want an immersive headset.

But when you’re on the go, you’ll want lightweight glasses that can give you timely information without reaching for your phone. We built Android XR together as one team with Samsung and optimized it for Snapdragon with Qualcomm. Since releasing the Android XR developer preview last year, hundreds of developers are building for the platform. We’re also reimagining all your favorite Google apps for XR, And it’s Android after all, so your mobile and tablet apps work too. Now today, I want to share how Gemini transforms the way you experience both headsets and glasses.

On these devices, your AI assistant understands your context and intent in richer ways to help you throughout your day. Let’s start with Gemini on headsets. This is Samsung’s Project Moohan, the first Android XR device. Moohan gives you an infinite screen to explore your apps by with Gemini by your side. With Google Maps in XR, you can teleport anywhere in the world simply by asking Gemini to take you there.

You can talk with your AI assistant about anything you see and have it pull up videos and websites about what you’re exploring. So many of us dream about sitting front row to watch our favorite team. Imagine watching them play in the MLB app as if you were right there in the stadium while chatting with Gemini about player and game stats. Samsung’s Project Mulhan will be available for purchase later this year. We can’t wait for you to try it for yourselves.

Android XR Glasses

Now let’s turn our attention to glasses. As you know, we’ve been building glasses for over ten years, and we’ve never stopped. Glasses with Android XR are lightweight and designed for all day wear even though they’re packed with technology. A camera and microphones give Gemini the ability to see and hear the world. Speakers let you listen to the AI, play music, or take calls.

And an optional in lens display privately shows you helpful information just when you need it. These glasses work with your phone, giving you access to your apps while keeping your hands free. All this makes glasses a natural form factor for AI, bringing the power of Gemini right to where you are. So unlike Clark Kent, you can get superpowers when you put your glasses on.

Okay. Who’s up for seeing an early demo of Android XR glasses? Let’s see how they work in the most hectic environment possible right now, backstage at IO. Our very own superwoman, Nishta, is back there to show us how these glasses work for real. Let me send her a text now, and let’s get started.

Live Demo of Android XR Glasses

NISHTA: Hey, everyone. Right now, you should be seeing exactly what I’m seeing through the lens of my Android XR glasses, like my delicious coffee over here and that text from that just came in. Let’s see what he said. Alright. It’s definitely showtime. So I’m going to launch Gemini and get us going.

Send Jerome a text that I’m getting started and silence my notifications, please.

GEMINI: Okay. I’ve sent that message to him and muted all your notifications.

NISHTA: Perfect. Oh, hey, Dieter.

DIETER: Hey, Nesha. I see the lights on on your glasses, so I think it’s safe to say that we’re live right now?

NISHTA: Yes. We’re officially on with the IO crew. Hey, everybody. It is pretty great to see IO from this angle.

DIETER: Nishta, you promised me I could get my own pair of Android XR glasses if I helped out back here. So, what do you say?

NISHTA: Of course. Let’s get coffee after this, and I’ll bring you those glasses.

DIETER: Awesome. We’ll see you then. Good luck.

NISHTA: Thank you. As you all can see, there’s a ton going on backstage. And is that pro basketball player, Janis, wearing our glasses?

JANIS: I love it. This fits on both of my hands for double high fives. Nice.

NISHTA: Let me keep showing you guys what these glasses can do. I’ve been curious about this photo wall all day. Like, what band is this and how are they connected to this place?

GEMINI: Band. No way. Can you show me a photo of one of their performances here?

NISHTA: Sure. Here’s one. Want me to play one of their songs?

GEMINI: I’d love that. I can listen while I make my way to the stage.

NISHTA: Great. Here’s Under the Aurora by Counting Crows.

SHARAM: Hey, everyone. Thanks for that staff studded behind the scenes look. By the way, do you want to book that coffee with Dieter now?

NISHTA: Yes. The crew actually gave me some awesome coffee backstage, so let me try something fun. Gemini, what was the name of the coffee shop on the cup I had earlier?

GEMINI: That might have been Bloomsgiving. From what I can tell, it’s a vibrant coffee shop on Castro Street.

NISHTA: Great memory. Can you show me the photos of that cafe? I want to check out the vibe.

GEMINI: Definitely. Do these photos from maps help?

NISHTA: Oh, I know that spot. It’s a flower shop as well as a coffee shop, but it is downtown. Okay. Gemini, show me what it would take to walk here.

GEMINI: Getting those directions now. It’ll take you about an hour.

NISHTA: Okay. I can get some steps in and these heads up directions and a full 3D map should make it super easy. Go ahead and send Dieter an invite for that cafe and get coffee at 3PM today.

GEMINI: I’ll send out that invite now. Enjoy the coffee.

SHARAM: As you saw, Gemini helped Nishta search what she sees, remember details like the coffee cup, book an event, even navigate, all without taking her phone out of her pocket. I’m even wearing the glasses right now too. They’re my personal teleprompter, and I have prescription lenses so I can see you all. Okay, Nishta. This is a big moment for glasses. Let’s capture it.

NISHTA: Yes. Get ready for a quick photo, everyone, and let’s bring out our star. It’s on screen shot. Alright, Gemini. Done it. Take a photo for me. Alright, Gemini. Take a photo for me and add it to my favorites. That looks amazing. I’m completely starstruck.

SHARAM: Okay. One last thing. Sundar showed what’s possible with live chat translation earlier. Let’s see what that’s like on glasses. This is a very risky demo, but we’re going to give it a shot. Nishta and I are going to speak to each other in our mother tongues. Nishta is going to speak Hindi. I’m going to speak Farsi very poorly. And you’ll see the feed from both of our glasses back here. And so you can all follow along, we’ll show an English translation in real time. Okay? Let’s give it a shot. Fingers crossed. We said it’s a risky demo.

NISHTA: Thank you so much.

SHARAM: Thank you, Nishta, for that awesome tour of Android XR glasses. We’re so excited about the possibilities when you have an incredibly helpful AI assistant by your side with these Android XR devices. But that’s not all. We’re taking our partnership with Samsung to the next level by extending Android XR beyond headsets to glasses.

We’re creating the software and reference hardware platform to enable the ecosystem to build great glasses alongside us. Our glasses prototypes are already being used by trusted testers, and you’ll be able to start developing for glasses later this year. Now we know that these need to be stylish glasses that you’ll want to wear all day. That’s why I’m excited to announce today that Gentle Monster and Warby Parker will be the first eyewear partners to build glasses with Android XR. We want you to be able to wear glasses that match your personal taste.

This is just the start. I can’t wait for you to try AndroidXR for yourself, and we’ll have lots more to share in the months ahead. Thank you so much.

Closing Remarks

SUNDAR PICHAI: Those six star classes are amazing. Research to reality, literally right in front of our eyes.

Okay. So we are nearing the end of the show. Today, you’ve heard a lot about ELO scores, benchmarks, and state of the art performance. But I know there’s one metric you’ve all been waiting for, our AI counter. So let’s take a look at one last leaderboard.

Looks like, I guess, we have a new entrant. Gemini takes the lead, coming in at 95. Very exhilarating. On a more serious note, here’s everything we’ve announced today, from new launches and product expansions to glimmers of what’s to come. The opportunity with AI is truly as big as it gets, and it will be up to this wave of developers, technology builders to make sure its benefits reach as many people as possible.

I want to leave you with a few examples that inspire me. The first is top of mind for those who live here in California and so many places around the world. So many of us know someone who has been affected by wildfires. They can start suddenly and grow out of control in a matter of minutes. Speed and precision can make all the difference.

Together with an amazing group of partners, we are building something called Firesat. It’s a constellation of satellites that use multispectral satellite imagery and AI aiming to provide near real time insights. Just look at the resolution. It can detect fires as small as 270 square feet, about the size of a one car garage. Our first satellite is in orbit now. When fully operational, imagery will be updated with a much greater frequency, down from every twelve hours today to every twenty minutes.

Feed is also of the essence in other kinds of emergencies. During Hurricane Hareen, Wing, in partnership with Walmart and the Red Cross, provided relief efforts with drone deliveries. Supported by AI, we were able to deliver critical items like food and medicine to a YMCA shelter in North Carolina based on real time needs. You can imagine how this could be helpful in disaster relief in other communities and we are actively working to scale up.

These are examples of ways AI is helping society right now. It’s especially inspiring to think about the research of today that will become reality in a few short years, Whether it’s building the next generation of helpful robots, finding treatments for the world’s deadliest diseases, advancing error corrected quantum computers, or delivering fully autonomous vehicles that can safely bring you anywhere you want to go. All of this is very much possible within not decades, but years. It’s amazing.

This opportunity to improve lives is not something I take for granted. And the recent experience brought that home for me. I was in San Francisco with my parents. The first thing they wanted to do was ride in a Waymo, like a lot of other tourists. I had taken Waymos before, but watching my father, who was in his eighties, in the front seat, be totally amazed. I saw the progress in a whole new life.

It was a reminder of how incredible the power of technology is to inspire, to awe, and to move us forward. And I can’t wait to see what amazing things we’ll build together next. Thank you.

Audience Reactions

“See infinite possibilities with flow. What manner of magic is that?”

“AI mode is the biggest revolution since the search engine was invented.”

“You tell Gemini exactly what you want built, and it builds it for you. You’re not coding the Gemini shadow.”

“AI is completely changing how we fight fires.”

“Generative media is expanding the boundaries of creativity.”

“It was magic the first time I saw this.”

“This isn’t just another tool. It’s a game changer.”

“This is just, like, mind blowing to me. The potential is almost infinite.”

Related Posts

• Full Transcript: Zanny Minton Beddoes Interviews Elon Musk
• Transcript of Daniel Kokotajlo Interview: Diary Of A CEO Podcast
• Transcript: Why a Social Media Ban Won’t Save Teens – Candice Odgers  
• Transcript of Elon Musk Interview: SXSW Live – March 9, 2013
• Digital Afterlife: Who Owns Your Data When You’re Gone? – Suresh Sankaran Srinivasan (Transcript)