Everything Google announced at I/O 2026: new Gemini models, AI agents, and Android XR glasses

Daria Ivanova — 20.05.2026

Google used its annual I/O 2026 conference to outline a wide-ranging set of updates that reached well beyond its developer audience.

Gemini sat at the center of the event, with new models, an agent-driven workflow, a redesigned app, and tighter integration across Search, Gmail, Docs, and shopping. Google also introduced its first consumer hardware running on the Android XR platform: AI-powered smart glasses set to launch this fall.

Here is a breakdown of the most important announcements, ordered from the features users can try right now to the experimental tools aimed at developers.

Gemini 3.5 Flash: a new model available to everyone

The headline technical announcement was the Gemini 3.5 family of models, designed around a single idea: combining frontier-level intelligence with the speed needed for real-time agent workflows.

The first model in the family, Gemini 3.5 Flash, is already rolling out for free across the Gemini app, AI Mode in Google Search, the Gemini API via AI Studio and Android Studio, and enterprise products such as Gemini Enterprise.

Flash outperforms Gemini 3.1 Pro on several key benchmarks:

  • Terminal-Bench 2.1 — 76.2%
  • GDPval-AA — 1656 ELO
  • MCP Atlas — 83.6%
  • CharXiv Reasoning (multimodal understanding) — 84.2%

The model is also four times faster than other frontier models at generating responses, which makes it especially well suited for agentic workflows where latency matters. It runs at less than half the cost of comparable top-tier models.

For scale, Google now processes more than 3.2 quadrillion tokens per month across its products, seven times more than at last year's I/O. The Gemini API alone handles roughly 19 billion tokens per minute.

Gemini 3.5 Pro is still on the way. The company said the model is already running internally and will launch publicly next month, but did not share any technical specifications.

Gemini Omni: a model that generates video

The second major model announcement was Gemini Omni, a new family that takes the leap from text and image generation to producing video as an output.

The idea is straightforward. Users feed in any combination of text, photos, and existing video, and the model returns a clip that respects context, follows physics, and keeps characters and details consistent from frame to frame.

Editing happens through normal conversation, with each new instruction layered on top of the previous one rather than replacing it. A user can ask the model to swap out a background, add a character, change the lighting, or transform an object, and Omni interprets each request in the context of the scene rather than applying a flat filter.

Several examples of what Gemini Omni can do:

One standout feature is action replacement: users can take an existing piece of footage and ask Omni to change what is happening in the frame, insert new characters or objects, or turn an ordinary moment into something that would have been impossible to capture on camera.

The first model in the lineup, Gemini Omni Flash, is available starting today to AI Plus, Pro, and Ultra subscribers across the Gemini app, Google Flow, and YouTube Shorts. Google plans to expand the family to include image and audio outputs as well, though no timeline has been set.

Google is positioning Omni not as another Sora or Runway rival, but as a model built to understand how the real world works — from gravity and lighting to natural human movement. The idea is simple: better grounding should make videos look more realistic and easier to edit.

A redesigned Gemini app: new design and Daily Brief

Google has fully rebuilt the Gemini app around a new design language called Neural Expressive, and the change runs deeper than the visuals.

The interface brings smoother animations, brighter colors, fresh typography, and tactile feedback. The input field is now a pill shape with a single plus button on the left that opens every tool. More significant is what happens inside each response. Gemini now decides how to format an answer on its own, weaving in images, interactive timelines, narrative videos, or dynamic graphics where they make sense, rather than returning a wall of text. The feature works with mixed results on Gemini 3.5 and does not run on Gemini 3.1 Pro at all. The most important information is pushed to the top of the response and highlighted in bold.

Gemini also suggests follow-up buttons to keep the conversation going.

The other major shift is that Gemini Live is now built directly into the main interface. Previously, voice mode opened in a separate full-screen window, which forced users to switch between modes. The redesigned version lets users move between text and live voice conversations seamlessly without losing context. Voice itself has been overhauled as well — the model no longer cuts users off mid-sentence, and people can speak at their own pace. Regional dialects are coming soon, with options to pick a voice that sounds closer to the user's own. Neural Expressive is rolling out globally on Android, iOS, and the web.

The second addition is Daily Brief, a personalized morning digest. The agent scans Gmail, Calendar, and Tasks, surfaces what matters most for the day ahead, and suggests next steps. It looks like this:

The brief can prompt the user to reply to an email, move a meeting, or set a reminder. It is available today to AI Plus, Pro, and Ultra subscribers in the U.S. only.

Gemini Spark: an agent that completes tasks for you

The biggest consumer announcement of the conference was also the clearest signal of where Google sees Gemini heading.

Gemini Spark is a personal AI agent that runs 24/7 in the background and actually completes tasks on the user's behalf rather than just answering questions. Google describes it as a shift from an assistant that responds to a partner that gets things done. Spark is wired into Gmail, Docs, Calendar, and the rest of the Google ecosystem, and can find a specific email, draft a document, schedule a meeting, or pull information together from multiple sources to deliver a finished result. This summer, support for the MCP protocol will let Spark work with third-party services as well, extending it beyond Google's own products.

A separate piece of the rollout is the macOS Gemini app, which will gain Spark integration so the agent can operate directly on the local machine. It will be able to run tasks and interact with files and apps on the computer, pushing Spark beyond the browser and mobile.

During the live demo, Google showed a user selecting several documents in a folder and then asking the agent, by voice, to write a friendly email and build a table from the data inside those files.

Alongside Spark, Google introduced Android Halo, a smaller but related feature. It is a thin bar at the top of the Android screen that shows the agent's status in real time: what it is doing, how far along it is, and whether there is a new message. The bar stays visible on top of any app, so users do not need to switch contexts to check. Android Halo arrives later this year and will work with Gemini Spark as well as other supported agents.

For now, Spark is in beta and limited to AI Ultra subscribers, on both the $100 and $200 tiers, in the U.S. only. A wider rollout is planned for next week. Google has not said when the feature will become available outside the U.S.

Google AI subscriptions: a new tier and a new limits system

Google has reshuffled its subscription lineup and added a new AI Ultra tier at $100 per month, aimed at developers, technical professionals, and power users. The new plan comes with limits five times higher than AI Pro, priority access to Antigravity, 20 TB of cloud storage, and a YouTube Premium subscription. The previous top tier has dropped from $250 to $200 a month, with the same capabilities intact. The $200 plan offers limits 20 times higher than Pro.

A more interesting change is how those limits are measured. Google used to count the number of queries. The new system counts compute: a simple text query uses very little, while a complex query involving video generation consumes much more. Limits refresh every five hours up to a weekly cap. Google argues this is a fairer model because it accounts for the gap between light and heavy tasks. Anthropic uses a similar system in Claude.

Gemini Spark is available across both Ultra tiers, $100 and $200, currently in the U.S. only. Gemini Omni is available to all Plus, Pro, and Ultra subscribers globally.

Google Search: agents, a smarter bar, and mini-apps

Search received what is arguably its most significant update in years, with Google calling it the biggest overhaul of the search bar in 25 years.

  • AI Mode now runs on Gemini 3.5 Flash and is globally available to all users. A year after launch, AI Mode has crossed a billion monthly active users, and query volume is doubling every quarter. The average AI Mode query is three times longer than a typical search query, as users increasingly write out detailed questions instead of relying on keywords.
  • A redesigned smart search bar. It dynamically expands as the user types and supports text, photos, files, video, and open Chrome tabs. AI suggestions work less like autocomplete and more like intent prediction, offering to refine or rephrase the query. Conversational context is preserved between queries, so users can ask follow-up questions straight from an AI Overview and slide into a full conversation with AI Mode.
  • Information agents. They monitor the web 24/7 on topics the user defines — news, finance, sports, listings, social media — and deliver synthesized updates at the right moment. An agent can be set to search for apartments based on specific criteria and notify the user the moment a matching listing appears, or to track collaborations from a favorite sneaker brand. Information agents will roll out to AI Pro and Ultra subscribers this summer.

Google also announced mini-apps inside Search — custom trackers and dashboards built for specific user tasks. They are also coming to Pro and Ultra subscribers in the coming months.

Universal Cart: one cart for everything

Google has consolidated shopping across its entire ecosystem into a single cart. Universal Cart works simultaneously across Search, the Gemini app, YouTube, and Gmail.

The moment a user adds an item, the cart starts working in the background — tracking prices, hunting for discounts, showing price history, and flagging when an out-of-stock item is available again. For multi-part builds, such as PC components sourced from different retailers, the cart checks compatibility and suggests alternatives if something does not fit.

Because Universal Cart is built on Google Wallet, it factors in payment methods, loyalty programs, and store promotions, recommending the best way to pay and where.

Checkout runs through Google Pay in a few taps at supported partner retailers. The initial launch partners include Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify stores such as Fenty and Steve Madden. Universal Cart launches in Search and the Gemini app in the U.S. this summer, with YouTube and Gmail to follow.

Google Workspace: voice instead of keyboard

The dominant theme of this year's Workspace updates was voice control. Google added conversational features to three products at once: Gmail, Docs, and Keep.

  • Gmail Live is a voice search tool for email. Users can ask out loud, "What's my flight's gate number?" or "What's happening at school this week?" and Gmail Live finds the answer in their messages and surfaces it without forcing the user to dig through folders.
  • Docs Live takes things further as a full voice-based co-author. Users can dictate a stream of thoughts, and Docs Live structures what was said into a document, pulling in relevant data from Gmail, Drive, and the web as needed. The document can be edited by voice, the tone adjusted, and details refined, all in live conversation. Down the line, Google plans to let users create new documents and edit existing ones entirely by voice.
  • Keep uses simpler logic but is no less useful. Users can dump thoughts aloud in any order, and the app turns them into structured notes and lists in the background.

All three features launch globally for AI Pro and Ultra subscribers this summer. In parallel, Google announced Google Pics, a new app for generating and precisely editing images with a high degree of control over the result.

AI Inbox, which helps users manage email, is also expanding to AI Plus and Pro subscribers, after previously being limited to AI Ultra.

Separately, Google introduced Ask YouTube, a smart search feature that understands complex queries, surfaces relevant videos, and jumps directly to the right moment in each clip. It is currently being tested with Premium subscribers.

Android XR: smart glasses this fall

Google announced the first hardware running on the Android XR platform: Gemini-powered glasses launching in the fall of 2026.

The platform was built in partnership with Samsung and Qualcomm, with two brands handling the actual designs: Gentle Monster and Warby Parker. The glasses will be sold as full collections, and crucially, they work with both Android and iPhone.

The interface activates with "Hey Google" or a tap on the temple. From there, users can ask Gemini about whatever they are looking at — restaurant reviews while walking past a place, an explanation of a confusing parking sign, or the name of a cloud overhead. Navigation works based on where the user is standing and where they are looking, with the glasses giving step-by-step voice directions and accepting requests to add a stop or find a nearby café. Calls, messages, and missed notifications are handled by voice through built-in speakers that Google describes as clear and private.

The glasses have a camera, which lets users take photos or record video by voice. Nano Banana can immediately suggest removing distractions in the background or applying an effect. They can also translate speech in real time, preserving the original speaker's tone and intonation.

Display glasses that show information directly in the user's field of view will follow later. Google has not announced a specific timeline.

SynthID and Content Credentials: telling AI from reality

Google is expanding its content verification tools, which have moved beyond experiment status and now operate as infrastructure across the entire ecosystem.

SynthID, Google's invisible watermarking technology for AI-generated content, has been around for three years. In that time, the company has watermarked more than 100 billion images and videos and 60,000 years of audio. SynthID verification is now expanding to Search, available today, and to Chrome in the coming weeks. Users will be able to ask "Is this AI-generated?" right from Search or the browser.

In parallel, the Gemini app now supports C2PA Content Credentials, a standard that shows whether a piece of content was captured by a camera or created and edited using AI. Support is coming to Search and Chrome in the coming months. OpenAI, Kakao, and ElevenLabs have joined SynthID — a meaningful step, since the technology only works at scale with broad industry buy-in.

For developers: Antigravity 2.0, AI Studio, and Stitch

  • Antigravity 2.0 is a new desktop application that serves as a control center for agents. It lets users run multiple agents in parallel, schedule background tasks, and integrate with AI Studio, Android, and Firebase. For users who prefer the terminal, Google also launched Antigravity CLI, and recommends migrating to it from Gemini CLI.
  • AI Studio added native Android app building directly in the browser, removing the need for SDKs or a local development environment. It also gained integration with Google Workspace, so users can build dashboards based on Sheets and work with Drive and Docs straight from AI Studio. A mobile version of AI Studio is open for pre-registration.
  • Google Stitch is a tool for designing interfaces in real time by voice or text. The result appears on the workspace as the agent works, so users can adjust each iteration before it is finalized. Finished designs can be exported to Antigravity or published through Netlify. The tool is globally available starting today.

Google Flow, Pomelli, and experiments

  • Google Flow, the company's AI studio for video creation, picked up Gemini Omni Flash and a new Flow Agent that helps at every stage of production. Mobile versions are also rolling out: Flow on Android (in beta, iOS soon), and Flow Music on iOS (Android soon). Flow Music runs on the Lyria 3 Pro model and is aimed at musicians and producers.
  • Pomelli, available in Google Labs, is a tool for small businesses. The agent helps build a brand identity, generate a brand book, and assemble a website in just a few clicks. It is in Google Labs now.
  • Project Genie, the experimental interactive world generator, can now tie generated spaces to real locations through Google Street View. It is rolling out to all AI Ultra $200 subscribers.
  • Gemini for Science includes three experimental tools for researchers in Google Labs: hypothesis generation, algorithm evolution, and research assistance, along with Science Skills in Antigravity.

The bigger picture

Google I/O 2026 was less a collection of separate announcements than a demonstration of a single, unified strategy. Gemini is no longer being positioned as a chatbot but as a layer running through Search, Gmail, Docs, shopping, creative tools, and now physical hardware.

The agent direction — Spark, the information agents in Search, Universal Cart — is an attempt to let users set a task once and have the system carry it through to completion. Whether the vision actually works in practice will become clearer this fall, when most of the announcements are scheduled to reach a wider audience.