Google I/O 2026: Everything Announced

Google I/O 2026 opened on May 19 at the Shoreline Amphitheatre in Mountain View with a clear message from Sundar Pichai: Google is in the agentic era now, and this keynote was built to prove it.

The 9.7 trillion tokens processed through Gemini every month was one of the first numbers Pichai put on screen. That scale is the backdrop for everything else announced today: Google is not experimenting with AI. It is deploying it at infrastructure scale across every product it ships.

I watched the full keynote and have been going through the technical details since. Here is every major announcement, what it actually means, and where the caveats are.

---

Gemini 3.5 Flash: Available Right Now

Gemini 3.5 Flash combines frontier intelligence with the ability to perform agentic tasks. It surpasses Gemini 3.1 Pro in coding, agentic, and multimodal benchmarks, with the cost and speed of the Flash series at 4x faster than other frontier models in terms of output tokens per second. It is rolling out starting today in the Gemini app, Search, Antigravity 2.0, and the Gemini API.

The benchmark numbers that matter: 90.4% on GPQA Diamond, a benchmark widely used to test PhD-level scientific reasoning, 81.2% on MMMU-Pro for multimodal understanding, and 78% on SWE-bench Verified in coding.

That last number deserves specific attention. 78% on SWE-Bench Verified from a Flash model is genuinely remarkable. To put it in context: Gemini 3.1 Pro scored 80.6% on SWE-Bench Verified when it launched in February. Gemini 3.5 Flash, which is designed to be faster and cheaper than Pro, gets within 2.6 points of that while running four times faster. This is the most significant capability-per-cost jump in the Flash tier since the series launched.

Gemini 3.5 Pro is currently in testing and will be available next month. Based on the trajectory of the Flash model, the Pro variant should challenge GPT-5.5 and Claude Opus 4.7 when it arrives.

For developers: Gemini 3.5 Flash is available via the Gemini API, Google AI Studio, and Vertex AI today. It is the model now powering AI Mode in Google Search, which means every search query going through AI Mode is now running on this model.

---

Gemini Spark: The 24/7 Personal Agent

This was the announcement that generated the most discussion in the room, and it deserves careful explanation because it is different in kind from what Gemini has done before.

Gemini Spark is a persistent AI agent that monitors Gmail, Calendar, Google Tasks, and connected third-party apps in the background without requiring the user to open the Gemini app. Demonstrated live on stage, Spark drafted and sent emails via voice on macOS, prepared morning briefings, and handled multi-step scheduling tasks.

The demo Google ran involved organizing a party: Spark created an RSVP tracker in Google Sheets, populated it with contacts from Gmail, sent invitations, and updated the tracker in real time as responses came in. All of this happened while the presenter was giving the demo without manually triggering each step.

Gemini Spark is described as "your personal agent" that takes actions on your behalf to help "navigate your digital life." Google says this represents "a big shift for Gemini, transforming it from an assistant that can answer your questions into an active partner that does real work on your behalf and under your direction."

Spark rolls out as a beta to trusted testers this week and to US Google AI Ultra subscribers, at both the $100 and $200 tiers, starting next week, with broader availability and third-party app integrations to follow.

One important note that most coverage is not highlighting: a leaked onboarding screen from an early Gemini app beta had drawn scrutiny in developer communities ahead of the event. That disclosure arrives alongside an existing proposed class-action lawsuit, Thele v. Google LLC, filed in November 2025 in federal court in San Jose, which alleges Google secretly enabled Gemini across all Gmail, Chat, and Meet accounts in October 2025 without user consent. Users considering enabling Spark can disable Gemini's access to Workspace apps through the Data and Privacy section of their Google Account settings.

If you plan to use Spark, review your Workspace data access settings first.

---

Gemini Omni: Any Input, Any Output

Demis Hassabis took the stage to introduce Gemini Omni, and this is where the technical ambition of I/O 2026 becomes clearest.

Gemini Omni is a new series of models that combines Gemini's reasoning capabilities with creation. Gemini Omni Flash today accepts image, audio, video, and text input and outputs video grounded in real-world knowledge and that can be easily edited.

The headline reveal was Gemini Omni, a next-generation multimodal AI architecture designed to process and generate multiple forms of media simultaneously. Unlike earlier AI systems that handled text or images separately, Gemini Omni combines visual understanding, voice interaction, video generation and reasoning together.

The capabilities demonstrated on stage: editing existing videos conversationally by describing what you want changed, generating new video clips from text prompts, taking a selfie and transforming it into a different visual style, and a demo of 16 simultaneous camera angles generated in parallel from a single scene using 16 independent agents through Gemini Flow.

Gemini Omni is available in YouTube Shorts Remix and the Create app. Google Flow and Google Flow Music are the new creation tools built on Omni.

The longer-term statement from Google is significant: Gemini Omni will work with videos first, but Google says it will be upgraded over time to generate any output from any input. That is not a small claim.

---

Google Search: The Biggest Redesign in 25 Years

Google is also combining AI Overviews and AI Mode into a more unified experience. Users will now be able to move seamlessly between traditional search results, AI-generated answers and follow-up conversations without losing context. According to Google, the deeper users go into conversations, the more relevant links and sources become.

The specific features announced:

A new intelligent Search box expands the more you type in a reflection of how people have longer and more conversational queries. AI-powered query suggestions go beyond autocomplete by anticipating your intent. New information agents work in the background 24/7 to keep you updated on whatever matters most to you, intelligently looking across everything on the web including blogs, news sites, social posts, plus real-time data on finance, shopping, and sports. Available to Google AI Pro and Ultra subscribers this summer.

Google Search will soon be able to build custom dashboards and trackers for continuous tasks, described as mini apps for specific tasks, coming in the coming months for AI Pro and Ultra in the US.

Google announced Universal Cart, an AI-powered shopping cart that works across Google services including Search and the Gemini app, with YouTube and Gmail support coming at a later time.

For anyone who has been skeptical about whether Google's Search AI features meaningfully improve the product: the background monitoring agents are the feature to watch. A search engine that watches for changes to things you care about and tells you proactively is a fundamentally different product than one you query manually. Whether it works as well in practice as it did in the demo remains to be seen.

---

Daily Brief: Your Morning AI Digest

Daily Brief is a personalized digest of the day ahead. Sifting through your Gmail, Calendar, and Tasks, it prioritizes and organizes what you need to do, while suggesting next steps. Rolling out to Google AI Plus, Pro, and Ultra starting today in the US.

This is a smaller feature than Spark but immediately available to a wider audience. If you use Google Workspace heavily, Daily Brief should be worth enabling today.

---

Android Halo: Agent Visibility on Your Phone

Android Halo provides at-a-glance visibility into what your agent is working on at any given time with subtle communication at the top of your phone screen, so you can see the agent's progress right from the top of any screen without having to stop what you are doing. Available later this year for Gemini Spark and other supported agents.

This is the UI layer that makes persistent background agents actually liveable. One of the friction points with autonomous AI agents is not knowing what they are doing without actively checking. Halo solves that with a non-intrusive status indicator that does not require switching apps.

---

AI Ultra at $100: The Pricing Shift That Matters

Google kicked off its annual developer conference with a keynote that carried an unusually direct message: the company halved the entry price for its top-tier AI subscription. The new $100 plan includes a 5x higher usage limit in the Gemini app than the existing $20 AI Pro tier, 20 terabytes of cloud storage, YouTube Premium, and, starting next week for US subscribers, beta access to Gemini Spark.

The previous $250 plan is now $200 with the exact same capabilities as before.

Alongside the price restructuring, Google announced that the Gemini app is moving away from daily prompt limits in favor of a compute-used model. Under that system, a simple text message consumes less of a subscriber's monthly allowance than a complex video-editing or coding request. When a subscriber reaches their cap on flagship models, the system shifts them to the faster, lighter Gemini 3.5 Flash model. Google will also allow subscribers to buy additional credits on a pay-as-you-go basis.

The $100 tier changes the competitive picture. Before today, the top tier AI subscriptions were: ChatGPT Pro at $200, Google AI Ultra at $250, Claude Pro at $20 (with Claude Max at $100-$200). Now Google Ultra is at $100, matching Claude Max and sitting below ChatGPT Pro. For users who want the latest Gemini features and already pay for YouTube Premium, the $100 price effectively bundles a lot of existing spend.

---

Android XR Smart Glasses: Real Hardware This Fall

At Google I/O 2026, Google announced new intelligent eyewear built with Samsung and Qualcomm, in frames from Warby Parker and Gentle Monster, shipping this fall. The audio-only frames pair with Android and iOS so a Gemini agent can run errands on your phone while you stay heads-up.

What Google and Samsung have not shared yet: prices, exact ship dates, or which specific styles in each brand's collection get the Gemini features beyond the two preview designs. Meta's Ray-Ban Display, the closest comparison on the market right now, starts at $799. Google said more details are coming in the coming months.

Android XR is the OS layer Google built with Samsung and Qualcomm to span the full range of extended-reality hardware, from cheap audio frames all the way up to mixed-reality headsets. This strategy involves one platform, multiple hardware partners, a ladder of price points, with Gemini baked in at every tier. XREAL Project Aura, a wired XR device running on Android XR, was teased separately for a launch by end of 2026.

The glasses demo involved a live walkthrough of Gemini providing real-time contextual assistance through the device without touching a phone. Whether this represents a genuine consumer product moment or Google Glass repeating itself is the question the fall launch will answer.

---

Ask YouTube: A Redesigned Discovery Experience

Ask YouTube entirely reimagines how you find videos. Rather than keyword search, you can ask natural language questions and receive AI-curated answers combining video clips, channel recommendations, and summarized information from across YouTube's content library.

This is directly competitive with how younger audiences already use AI tools as an alternative to traditional search. Google is integrating the capability natively into YouTube rather than fighting the trend.

---

What Is Coming Next

Several announcements were previewed but not fully launched today:

Gemini 3.5 Pro is in testing and expected next month. Based on the Flash model's benchmarks, the Pro variant should be meaningful.

Android 17 is expected to release imminently. The final beta shipped in mid-April. Full Android 17 integration with Gemini Intelligence, the autonomous on-device agent layer, follows the stable release.

Android Halo for Gemini Spark and third-party agents arrives later this year alongside Spark's broader rollout.

Universal Cart third-party integrations including YouTube and Gmail come after the initial Search rollout.

Gemini Omni Pro, the full-capability version of the Omni model series, follows at a later date.

Broader Gemini Spark availability beyond the US comes after the initial Ultra rollout, with third-party app integrations expanding over time.

---

My Take on What Actually Matters

I/O 2026 was the most coherent keynote Google has given in years. Everything points in the same direction: Gemini as the operating layer for everything Google ships. The message from Sundar Pichai was not "here are some AI features." It was "the agentic era starts today."

The feature with the most immediate practical impact is Gemini 3.5 Flash, available right now in the API. The benchmark numbers are strong, the speed advantage over Gemini 3.1 Pro is significant, and developers can start building on it today. If you have been using Gemini 3.1 Pro in production, test 3.5 Flash against your workloads immediately.

The feature with the most long-term significance is Spark. A persistent background agent that works across your entire Google account without you having to trigger it is a genuinely different category of product from anything Google has shipped before. The privacy questions are real and worth understanding before enabling it. The capability, if it works as demonstrated, changes how people will use Google services day-to-day.

The feature with the most uncertainty is the smart glasses. The Warby Parker and Gentle Monster branding is smart positioning, the Android XR platform gives it a hardware ecosystem, and the live demo was impressive. But Google has announced wearable AI hardware before. Fall 2026 will tell us if this is different.

Google is looking to spend 6x more on AI in 2026 than it did back in 2022, estimating a total of $190 billion in capital expenditures for 2026 alone. The scale of that investment makes the ambition of today's announcements legible. Google is betting its next decade on making Gemini the intelligence layer for everything it ships.

---

Compare Gemini 3.5 Flash Against Other Frontier Models

Wondering how Gemini 3.5 Flash stacks up against GPT-5.5, Claude Opus 4.7, and DeepSeek V4-Pro on the benchmarks that matter for your workflow?

Compare AI models on Renovate QR

The /tools directory is updated as new benchmark data arrives. Gemini 3.5 Flash benchmarks will be added as independent evaluations come in over the next week.

---

Published May 19, 2026, the day of Google I/O. All announced availability and pricing reflects information from the keynote and official Google blog posts. We will update this article as Gemini 3.5 Flash and Gemini Spark roll out to users and independent benchmark data becomes available.