Okay so Google I/O is in about two weeks and I’ve been following the leaks and session schedules pretty closely. This year feels different. Not in a vague “big things are coming” way. In a very specific “Google is trying to win back the narrative” way.
OpenAI dropped GPT-5.5 last month. Anthropic shipped Claude 4.5. Chinese labs are releasing open-weight models that run on consumer GPUs and actually compete with the big proprietary models. Google has been relatively quiet through all of this. May 19 is when they answer.
Here’s everything we know and expect to see.
The Event Itself
Google I/O runs May 19 and 20, with livestreamed keynotes and sessions open to everyone online. The main Google Keynote kicks off at 10:00 am PT and runs until 11:45 am. The Developer Keynote follows at 1:30 pm PT.
If you want to watch live, go to io.google/2026. The keynotes are free to watch from anywhere.
Gemini 4 — The One Everyone Is Watching
This is the announcement that actually matters for the AI race.
Gemini 4 is reportedly scoring 84.6% on ARC-AGI2 — a benchmark that measures genuine reasoning ability, not just memorization. That number is significant. It puts Gemini 4 in genuinely competitive territory with the best models from OpenAI and Anthropic.
Gemini Nano 4 was previewed on April 2 and will be a centerpiece of the event. It comes in two variants — Fast and Full — based on Gemma 4, promising 3x speed gains over previous Nano models. Google is positioning this as the foundation for on-device AI across Android, which means AI features that work without an internet connection on your phone.
Project Astra — Google’s pitch for a universal AI assistant — may also get an update at the event. Last year’s demo was impressive but limited. This year they need to show it actually works in real-world conditions.
Agentic AI and Coding Tools
The first developer keynote on May 19 focuses specifically on “agentic coding” — AI tools that handle routine development tasks so engineers can focus on architecture and product strategy.
This is Google’s direct response to Cursor. The AI coding tool market has been dominated by Cursor and GitHub Copilot and Google needs to show that its own stack can compete. Google’s AICore Developer Preview already supports tool calling, structured output, system prompts, and thinking mode. The I/O demos will show whether it can compete with real developer workflows.
For anyone building apps or working in tech — this session is worth watching. If Google’s agentic coding tools are actually good, it changes the options available to developers significantly.
AI Glasses — The Hardware Nobody Expected Google to Get Right
Google has partnered with Warby Parker and Gentle Monster for smart glasses powered by Android XR — similar to Meta’s Ray-Ban partnership.
I’ll be honest, I was skeptical about this one. Google Glass was one of the biggest product failures in tech history. The glasses were ugly, expensive, socially awkward, and the AI wasn’t useful enough to justify any of that.
What’s different this time: the hardware is reportedly under 50 grams — light enough to wear all day — and it has iOS compatibility, meaning it’s not limited to Android users. The Warby Parker partnership also means these will look like actual glasses instead of a science experiment strapped to someone’s face.
Whether this becomes a mainstream product depends entirely on what the Gemini-powered features can actually do in real conditions. The demo at I/O will tell us a lot.
Android 17
Google has been beta testing Android 17 since February. New features include the ability to remap gaming controllers natively for the first time on Android, and the ability to run apps in “bubbles” that you can minimize on screen until you’re ready to use them.
The event also includes a session on Android 17’s “Adaptive Everywhere” vision — where users move fluidly between phones, cars, living rooms, and immersive environments.
No dedicated “Android Show” this year like last year, so all Android news will be in the main keynote.
Aluminum OS — Android and ChromeOS Becoming One Thing
Google is planning to merge ChromeOS and Android into a unified platform, referred to as Aluminum OS.
This has been rumored for years. The idea is to design a real desktop version of Android — not just Android scaled up to a larger screen, but something actually built for desktop use. If Google shows meaningful progress on this at I/O, it’s a big deal for anyone who uses Chromebooks or is thinking about alternatives to Windows and macOS.
Boston Dynamics and Robotics
This is the one that genuinely surprised me when I dug into the I/O schedule.
Google DeepMind and Boston Dynamics announced a partnership at CES 2026 to integrate Gemini Robotics models into the Atlas humanoid robot. Atlas can now understand natural language commands, break high-level tasks into subtasks, analyze its environment through sensors, and execute autonomously. Hyundai — Boston Dynamics’ parent company — plans to manufacture up to 30,000 humanoid robots annually by 2028.
Google is going to demo this at I/O. A humanoid robot running on Gemini, taking instructions in plain English, doing physical tasks autonomously. That’s not a future product. That’s apparently ready to show right now.
Apple and Google Are Apparently Working Together Now
Apple and Google have entered a multi-year collaboration where the next generation of Apple Foundation Models will be based on Google’s Gemini models and cloud technology. Apple is using a process called model distillation to create smaller, efficient AI systems based on Gemini that run locally on iPhones, iPads, and Macs.
This one is strange. Apple and Google are competitors in almost every meaningful way. But on the AI model side, Apple apparently decided it makes more sense to build on Gemini than to develop everything from scratch. If this gets confirmed at I/O it’s one of the more surprising partnerships in recent tech history.
Veo 4 and AI Media Generation
Veo 4 video generation is expected, with 30-second clips. Updates to Nano Banana for AI images and Lyria for AI music are also expected.
Google’s Veo has been competitive with OpenAI’s Sora and this update should push it further. For creators and marketers, AI video generation is becoming a genuine tool rather than a novelty.
Why This I/O Matters More Than Usual
Here’s the context that makes all of this significant.
OpenAI dropped GPT-5.5 on April 23. Anthropic iterated Claude recently. Alibaba shipped Qwen on April 22. Chinese labs are delivering open-weight models that run on consumer GPUs and match proprietary models on coding tasks.
Google has been responding with previews and announcements but hasn’t had a major public moment to put it all together. I/O on May 19 is that moment. They’re not just announcing one product. They’re showing the entire stack — models, hardware, operating systems, robotics, developer tools — all at once.
It’s a show of force designed to make the market understand that Google’s AI capabilities extend far beyond any single product.
Whether it lands depends on the demos. Announcements are easy. Live demos that actually work are where these things get won or lost. Remember the Bard demo in 2023 that wiped $100 billion off Alphabet’s market cap because it got a question wrong? Google definitely remembers.
May 19 is two weeks away. I’ll be watching live and covering everything that actually matters here. Come back then.