Google I/O 2025: Gemini AI, Android XR glasses, Google Beam, Project Astra, Chrome and deep search news

The keynote speech has wrapped up after a whirlwind of AI announcements across Gemini, Search and Android XR.

Gemini, which Google wants to turn into a universal AI assistant, is now testing out a new “agent mode” that allows users to delegate complex tasks to the model, including apartment hunting and even replying to friends’ messages.

Aside from AI-powered advancements in video conferencing and coding, Google also introduced the latest versions of its advanced video and image generation models, as well as a new all-encompassing filmmaking tool called Flow.

The conference devoted significant time to a slew of updates coming to Google Search, which can now accomplish more complex tasks in “AI mode.” And the company closed the keynote by teasing its AI-powered Android XR glasses in a flashy live demo.

Google Glass is sort of back, I guess?

The company unveiled Android XR through a flashy demo where a Google employee wore the new glasses and walked around backstage of the I/O event, demonstrating examples of the Gemini-powered "Clark Kent" glasses projecting information onto a user's frames as they interact with the real world.

An example: recalling the coffee cup logo from a drink she held earlier — and asking Gemini to schedule a meeting at that coffee shop at a later date. The glasses were able to pull up directions and later take a photo as well. The glasses also did real-time language translation (with an onstage glitch, which the team acknowledged was a risk!).

The demo also included a brief cameo from Greek basketball player and Milwaukee Bucks (for now) superstar Giannis Antetokounmpo, who is on-site today.

Following last year's I/O event, Google notably caught some heat after its brand-new Gemini-powered AI search features offered some incorrect and bizarre answers. Some screen grabs purported to show Gemini recommending that people eat "one small rock a day" or add nontoxic glue to stop cheese from sliding off a pizza.

With Google moving full steam ahead on building out its AI functions in search, VP of search Liz Reid acknowledged last year's issues and described them as "edge cases and quite rare." Talking to reporters before the I/O event this year, Reid said the company has since improved quality, accuracy and factuality of its AI modes.

"We also still took them very, very seriously and made a lot of improvements since then, both upgrading to newer models and further just engineering work on how to tune and teach the model," Reid said.

Also launching today is a new AI filmmaking tool called Flow, which Google said is built for creators.

It combines capabilities from Veo, Imagen and Gemini. Users can upload or generate images, customize them with text prompts and turn them into clips, cobbling together a full-length AI video.

Google’s SynthID function embeds invisible watermarks into AI-generated media. Since its launch two years ago, SynthID has watermarked more than 10 billion pieces of content.

Now, the company is expanding its partnerships in an effort to watermark more content. And its newest SynthID detector, rolling out today to early testers, aims to make it easier to detect these watermarks in images, audio tracks, text and video.

One running theme at this keynote is that the suite of Google products that you use on an everyday basis are getting some sort of Gemini-powered AI functionality. That includes the web browser Chrome.

Google says it is unrolling a "browsing assistant" in Chrome so that as you're searching for campsites, you can ask Gemini follow-up questions like "Which campsites have river access?"

The function is going to be available to Google AI Pro subscribers in the U.S. this week.

Google’s newest video generation model, Veo 3, is now available. The model comes with native audio generation, meaning it can be prompted to create anything from sound effects to background sounds to dialogue.

“We’re entering a new era of creation with combined audio and video generation that’s incredibly realistic,” Google VP Josh Woodward said.

The company’s latest image generation model, Imagen 4, is also coming to the Gemini app today.

With the flurry of AI announcements today, Google also insists that anyone with concerns about these tools will have the option to choose what functions they connect to.

Speaking to reporters prior to the keynote, CEO Sundar Pichai said the company will give users controls to "turn off these features at any time."

Although he did not specify which specific features and functions, Pichai said users will have transparency on privacy settings.

"I think we are going to make these explicit user actions, and none of this will work ever automatically," Pichai said.

Much of the keynote today has been focused on AI-powered advancements to the company's search engine.

With the rise of generative AI, questions have surfaced about whether AI chatbots and agents will replace the traditional Google Search — but Google makes it clear it’s fighting hard to prevent search from going obsolete.

Through a custom image generation model trained specifically for fashion, Google now offers visual try-ons through Search.

“Our try-on experience works with your photo. It’s not some pre-captured image or a model that doesn’t look like you,” said Vidhya Srinivasan, Google’s advertising and commerce VP. “And then when it comes to clothes that you are interested in, the AI model is able to show how this material will fold and stretch and drape on people.”

Then, she said, users can purchase the item of clothing using Google’s new agentic checkout feature.

Users can now hop on a video call with Google through the company’s new Search Live function, which can watch through the phone camera and guide a user through real-world tasks like gardening, conducting a chemistry experiment or removing a stain from a shirt.

A Google spokesperson says 5,000 attendees are at this year's I/O conference. If you told me five years ago that 5,000 people would be cheering for a smarter search bar, I would have said you were crazy. But in a tech world where chatbots are the way that most people interact with AI, search bars are where the party's at.

By the way, if anyone was wondering about the name: "I/O" is a reference to a googol, the number that lent the tech giant its name and represents a one followed by 100 zeroes.

Google shared that its AI overviews, or AI-generated summaries of search queries, have become “one of the most successful launches in search in the past decade.”

The feature faced initial mockery in the weeks after launch, when users found misinformed or nonsensical answers during the AI tool’s experimental phase. But since then, these overviews have scaled up to more than 1.5 billion users every month in more than 200 countries and territories, the company said.

Now, Google is introducing an “AI mode” in the U.S. to transform Google Search. Users can ask more complex queries that trigger searches with more advanced reasoning, using what the company calls a “query fan-out technique.” It can also give personalized suggestions based on past searches, as well as connect to personal apps like Gmail for more context. The optional feature is slated to launch this summer.

Also coming to Google’s AI mode is Deep Search, which can issue hundreds of searches at a time.

After kicking off the keynote with a flashy AI-generated countdown video, Hassabis says the company's video generator (called "Veo") has a better understanding of physics, gravity, light and material.

The company showed examples of a bedsheet flying in the wind and a robot walking through water to demonstrate the generator's ability to simulate more natural environments. Google says these functions serve business purposes too, and can instruct robots in how to interact in real life.

"Understanding the physical environment will also be critical for robotics," Hassabis said.

Veo is a competitor in an increasingly competitive generative AI space, with OpenAI's Sora representing one of its biggest rivals.

Google wants Gemini to become a “world model,” which DeepMind CEO Demis Hassabis describes as a model that can “make plans and imagine new experiences by simulating aspects of the world just like the brain does.”

It would be a key milestone in the race to develop artificial general intelligence, he said, envisioning the model as a universal AI assistant that can understand context and take action on the user’s behalf.

"A universal AI system will perform everyday tasks for us," Hassabis said. "It will take care of mundane admin and surface delightful new recommendations, making us more productive and enriching our lives."

As vibe coding grows increasingly popular among everyday AI users, Tulsee Doshi, product strategy lead for Gemini, showcased how Google AI Studio can help users code just by inputting text directions.

Google’s newest asynchronous coding agent, Jules, is also now available in public beta.

“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,” Doshi said.

In an introduction of Gemini’s new AI-powered “smart replies,” Pichai showcased the AI assistant a crafting personalized reply giving advice to a friend who asked the user how they planned a past road trip.

The feature allows Gemini to automatically generate a reply based on things like the user’s notes in Google Drive and past reservations found in Gmail. It can also apparently replicate the user’s tone.

“Now, if I’m being honest, I would probably reply something short and unhelpful,” Pichai said. “But with my smart replies, I can be a better friend.”

The latest frontier for AI companies is "agentic" applications, where you can basically commission an AI app to gather information and carry out tasks for you over time. Google's "Project Mariner" hopes to do this by adding functionality in the Gemini app for things like searching for an apartment.

Pichai showed a demo where you can ask Gemini to look for an apartment for three people in Austin, Texas, with parameters like monthly budget, and amenities like a washer/dryer in building. Gemini then combed Zillow to find listings and can continue to browse for new listings.

"It's great for companies like Zillow, bringing in new customers and improving conversion rates," Pichai said.

An "experimental version" of the agent mode in Gemini will be "coming soon" for subscribers, Pichai said.

Highlighting Gemini’s ability to understand its real-world surroundings through a phone’s camera and screen-sharing capabilities, Google played a video showing the AI assistant correcting a user who mistakes a garbage truck for a convertible, a street light for a building and a utility box for a package.

“Why is this person following me wherever I walk?” the user asked.

“No one’s following you,” Gemini responded. “That’s just your shadow.”

Pichai showed off a demo of a new feature in Google Meet: live speech translation. Using an example of a person speaking with a vacation rental owner in South America, the demo showed real-time live translation from English to Spanish (and vice versa) over audio. It sounded pretty solid (although, I don't speak Spanish!).

English and Spanish translation is available now "for subscribers," with more languages rolling out "in the next few weeks."

The company just revealed its newest product, Google Beam, which Pichai described as an AI-first video communication platform.

Using multiple cameras to capture different angles, Beam renders 2D video streams into 3D visuals that aim to give a “more natural and deeply immersive conversational experience.”

The conference kicked off with a video showcasing the abilities of Google’s video- and image-generating models Veo and Imagen, before CEO Sundar Pichai walked onstage with opening remarks about Gemini and its capabilities.

“I learned that today is the start of Gemini season,” Pichai joked. “Not really sure what the big deal is. Every day is Gemini season here at Google.”

Aside from its benchmark performance and coding abilities, Pichai noted that Gemini was able to complete the Pokémon Blue video game — "bringing us one step closer to achieving API: artificial Pokémon intelligence."

T-minus 15 minutes to the keynote and American singer and artist Toro y Moi is crushing a pretty solid DJ set, intermittently asking the crowd if they're awake yet.

In a brief acknowledgement of the tension between AI and artists as of late, Toro y Moi told the crowd he sees AI as the future, but reminded attendees to have real people interactions too: "Shake some hands. Meet some people."

(As a fan, I would highly recommend checking out his 2022 album Mahal, as well as his sick set from Trona.)

Ahead of the Google I/O keynote, some analysts say this is the tech giant's moment to try to punch up at its competitors.

"Google needs to show the vision and create a framework that gains more mindshare with developers as OpenAI is at the top to the mountain. A key event for Google on AI and showing investors 'Don’t forget about us,'" said Dan Ives, managing director at Wedbush Securities.

Ives said it's "huge stakes" for Google in an AI arms race.

Beautiful morning here in Mountain View, California (I'm sweating already!). The DJs have a pretty nice warm-up setlist going (Anderson .Paak and Kaytranada are heavily featured), and the Google employees organizing the event are busting moves at the entrance. There's a lot of buzz with people filing into the keynote speech, which is going to kick off here at the Shoreline Amphitheatre in about 30 minutes.

Like many Silicon Valley events these days, phone cameras are being pointed in every which way, complicated rigs and gimbals are flying through crowds and taking shots. Big tech loves influencers, and there appear to be many on-site today!

Gemini is Google's flagship large language model, a competitor to others like OpenAI's ChatGPT and Meta's Llama.

Its latest Gemini 2.5 model, released in March, is the company's most intelligent yet, capable of undergoing a more complex "reasoning" process before giving a response.

According to Artificial Analysis' independent AI ranking, Gemini 2.5 Pro outperforms competitors from xAI, Anthropic and Meta, as well as high-performing Chinese models from DeepSeek and Alibaba. Currently, it trails only behind OpenAI's o4-mini.

CNBC’s Deirdre Bosa joins "Money Movers" to discuss how Google is planning to use AI to balance search. Further details are expected at its AI event today.

Hearing is believing? Not these days.

Rapid advancements in artificial intelligence have made it easy for just about anyone to generate convincing fake audio of people speaking.

Can you tell real from fake audio? We’ve assembled nine examples to test your ear. Don’t worry — even advanced software still has trouble telling real from fake.

Click here to take our AI audio quiz.

On the morning of Google's I/O conference, Apple announced that its own annual Worldwide Developers Conference will be taking place June 9-13 this year.

Similarly yesterday, Microsoft kicked off its annual Build conference, where it highlighted the world's advancement into the "era of AI agents."

"We envision a world in which agents operate across individual, organizational, team and end-to-end business contexts," Microsoft wrote in a blog post. "This emerging vision of the internet is an open agentic web, where AI agents make decisions and perform tasks on behalf of users or organizations."

Google yesterday launched its popular AI research assistant NotebookLM as a mobile app.

The tool can synthesize lengthy text documents into podcast-style summaries and analyses, with users having the ability to ask questions and interact with the AI podcast hosts.

Google will be livestreaming its I/O keynote today at 1 p.m. ET.

That video stream will be available on Google's website as well as YouTube.

Jobs traditionally done by women are more vulnerable to the impact of artificial intelligence than those done by men, especially in high-income countries, a report by the United Nations’ International Labour Organization showed today.

It found 9.6% of traditionally female jobs were set to be transformed compared with 3.5% of those carried out by men as AI increasingly takes on administrative tasks and transforms clerical jobs, such as secretarial work.

Human involvement will still be required for many tasks — and roles are more likely to be radically changed rather than eliminated, the report said.

Jobs in the media, software and finance-related roles are also at the forefront of change as generative AI expands its learning abilities.

Read the full story here.

Pop quiz! Can you spot the AI-generated art? Click below to give it a shot:

The artificial intelligence (AI) boom has brought with it a cornucopia of jargon — from “generative AI” to “synthetic data” — that can be hard to parse. And as hard as it is to really understand what AI is (see our explainer for that) having a working knowledge of AI terms can help make sense of this technology. 

As part of our series explaining the basics of AI, we created a short glossary of terms that will hopefully help you navigate the rapidly developing field.

Click here to check out our AI glossary.

Not long ago, Silicon Valley was where the world’s leading artificial intelligence experts went to perform cutting-edge research. 

Meta, Google and OpenAI opened their wallets for top talent, giving researchers staff, computing power and plenty of flexibility. With the support of their employers, the researchers published high-quality academic papers, openly sharing their breakthroughs with peers in academia and at rival companies.

But that era has ended. Now, experts say, AI is all about the product.

Just a day after he took office, President Donald Trump announced a high-dollar AI infrastructure project, dubbed Stargate, with joint investments from OpenAI, Oracle and SoftBank.

Early this year, OpenAI announced a partnership with the U.S. government on scientific research and nuclear weapons security. The company has also urged the federal government to make it easier for AI companies to train on copyrighted material, citing a need to “strengthen America’s lead.”

Trump has also shared multiple AI-generated images on his own social media platforms, but yesterday, he signed into law a major piece of AI regulation: a bill cracking down on deepfake pornography, which is often produced using AI tools.

Alphabet and Google CEO Sundar Pichai posted a picture yesterday of himself and Demis Hassabis, the CEO and co-founder of Google's DeepMind AI lab, ahead of today's event.

In January, a small lab in China took Silicon Valley by surprise when it unveiled a powerful AI model, DeepSeek, that outperformed top U.S.-developed models despite having been built on a fraction of the resources. Tech stocks plunged that week as DeepSeek shot to No. 1 in the Apple App Store.

Trump called China’s powerful DeepSeek assistant a “wake-up call for our industries that we need to be laser focused on competing.” Entrepreneur Marc Andreessen similarly described DeepSeek's R1 model as “AI’s Sputnik moment.”

Its launch sparked both panic and intrigue among tech executives and spurred renewed American eagerness to invest in what some now view as a high-stakes technological battle.

Building off of Gemini, Google’s flagship large language model, the company has been developing a research prototype called Project Astra, which aims to develop a “universal AI assistant” that can hold real-time conversations, remember old ones and see the world outside of itself.

“Agentic AI” is likely to be the next frontier, with the big tech companies offering bold visions of a future in which aspects of both work and life can be automated through virtual assistants.

If Google's going to unveil anything super cool, it's going to be here at I/O.

From Google TV to its Pixel phones, the I/O conference has historically been the venue for big announcements and major product unveils.

Recall that in 2012, co-founder Sergey Brin did a splashy demo of Google Glass, the wearable tech that flopped because of its high price and limited capabilities. The project actually existed until 2023, when Google stopped selling its enterprise glasses. But it's possible it could dip its toes back into glasses again, as speculation builds that it could expand on a partnership with Samsung to bring Android right to your very eyes.

For the past two years, Google’s annual I/O conference has been dominated by talk of AI, leaving its Android products on the sidelines. This year’s I/O is widely expected to be more of the same. 

The company might unveil announcements related to Gemini 2.5, Project Astra or Android XR, the yet-to-launch platform for Google's extended-reality headsets and glasses.

Google last week even gave Android its own virtual event, Android Show: I/O Edition, signaling that the larger two-day event will be freed up for more AI-related announcements. But it also unveiled new updates about phones’ built-in Gemini assistant in its Android showcase.

Good morning from Silicon Valley! Google's flagship I/O conference is set to kick off at 1 p.m. ET/10 a.m. PT from the Shoreline Amphitheater in Mountain View, California. There will also be a developer-specific keynote at 1:30 p.m. PT, with some demos in between. Throughout the day there will be some breakouts for specific platforms like Android, Chrome and Google Cloud.

Also hearing there could be some sort of extended reality announcement...