If you thought last year’s Google I/O was ambitious, Google CEO Sundar Pichai just walked onto the Shoreline Amphitheatre stage in Mountain View and casually announced the next era of computing. Google I/O 2026 agentic AI announcements confirmed what many had suspected: Google is no longer building features around AI. It is rebuilding everything around it. From a 24/7 personal AI agent to a multimodal model that generates video from virtually any input, the sheer breadth of what landed at I/O this year would make anyone’s head spin.
Welcome to the agentic Gemini era. Buckle up.
In This Article
The Numbers Are Staggering
Before getting to the products, the scale matters. Two years ago, Google was processing 9.7 trillion tokens a month across its surfaces. At last year’s I/O, it had grown to roughly 480 trillion. Today, that figure has jumped sevenfold to over 3.2 quadrillion tokens per month. Over 8.5 million developers are now building apps with Google’s models monthly, and the company’s model APIs are processing roughly 19 billion tokens per minute.
These are not vanity metrics. They represent real adoption at a speed that the industry has rarely seen before.
Read Also: Google I/O 2026: Every Major Announcement Expected at Today’s Keynote
Gemini 3.5 Flash: Frontier Speed at Half the Price
The headline model of the day is Gemini 3.5 Flash. According to Google, it outperforms the older Gemini 3.1 Pro across almost all benchmarks, with particular leaps in coding tasks. The really striking number is speed. Google says 3.5 Flash is four times faster than comparable frontier models when measured by output tokens per second, making it simultaneously one of the most capable and most nimble AI models available right now.
The cost angle is just as compelling. Google pointed out that companies shifting 80% of their workloads from other frontier models to 3.5 Flash could save over one billion dollars annually. That is not a rounding error. Gemini 3.5 Flash is available today across Google’s products and APIs, with Gemini 3.5 Pro following next month.
Gemini Spark: Your AI Agent That Never Sleeps
The product generating the most excitement is Gemini Spark, a personal AI agent baked into the Gemini app that operates 24 hours a day, seven days a week, on dedicated virtual machines in Google Cloud. You do not need to leave your laptop open. Spark is powered by Gemini 3.5 and the new Google Antigravity platform, designed to handle long-horizon tasks quietly in the background while you get on with your life.
It integrates with Google’s own tools from the start, with third-party MCP support arriving in the coming weeks. On Android, users will be able to monitor task progress through a new UI layer called Android Halo, launching later this year. Later this summer, Spark will operate directly within Chrome, acting as a fully agentic browser companion.
The Gemini Spark beta is rolling out to Google AI Ultra subscribers in the US next week.
Gemini Omni: When AI Learns to Dream in Video
Google also introduced Gemini Omni, a new model capable of generating output in any modality from any input. The first launch, Gemini Omni Flash, starts with video outputs and will expand to images and text over time. It fuses Gemini’s language intelligence with Google’s generative media models into something that feels genuinely new. You can try it today in the Gemini app, Google Flow, and YouTube Shorts. API access for developers and enterprises is coming in the next few weeks.
Ask YouTube and Docs Live: AI Meets Everyday Products
Google did not stop at model announcements. Ask YouTube entirely reimagines how users discover content on the platform. Instead of scrolling through search results and hoping for the best, Ask YouTube surfaces the most relevant moment in a video, surfaced specifically for your question. It begins US testing this summer.
In Google Docs, a new feature called Docs Live lets users verbally brain dump their thoughts and have Gemini organise, structure and write from that raw audio. No more wrestling with a blank page. Docs Live rolls out to subscribers this summer, with similar voice capabilities heading to Gmail and Keep at the same time.
Read Also: NYT Strands Hints and Answers for May 19, 2026 (Puzzle #807)
Search Gets an Agentic Upgrade
Google Search itself is entering the agentic era in a meaningful way. Information agents in Search are personalised AI agents working in the background around the clock to find information and trigger actions at exactly the right moment. These agents roll out this summer, initially for Google AI Pro and Ultra subscribers.
Search will also gain generative UI capabilities, building custom layouts and interactive visuals for individual queries. These tools will be free for everyone this summer, with persistent custom dashboards and trackers coming first for Pro and Ultra subscribers in the US.
SynthID Now Has OpenAI as a Partner
In a surprising cross-industry development, Google announced that OpenAI, Kakao and Eleven Labs are all adopting SynthID, Google’s invisible AI watermarking technology. Since its launch three years ago, SynthID has watermarked over one hundred billion pieces of content. Google is now expanding Content Credentials and SynthID verification into Search and Chrome, giving users an easy way to verify whether content was created by AI or a camera.
The Infrastructure Behind It All
To support this AI wave, Google’s capital expenditure is expected to hit between $180 billion and $190 billion this year, roughly six times its 2022 figure. The company also announced its eighth generation of TPUs, with a dual chip approach: TPU 8t for training, which offers nearly three times the raw compute of the previous generation, and TPU 8i for inference, optimised for speed and efficiency. Both deliver up to two times better performance per watt.
Google I/O 2026 in One Sentence
If this year’s I/O had a thesis, it was this: AI should not be a destination you visit. It should be the infrastructure you live inside. Google has made a very bold, very expensive, and frankly very coherent bet that agentic AI is the next computing paradigm. Whether Gemini Spark becomes as indispensable as Google Search once did remains a genuinely interesting question. But after today, dismissing the possibility feels considerably harder than it did yesterday.
