Google spent years building AI into its products. At I/O 2026, it made clear the next chapter isn’t about features — it’s about agents that work while you don’t.
As every year, Google has introduced many new updates for developers and designers. At Google I/O 2026, they shared how they’re making AI more helpful for everyone. Â
At the conference, Google released two new models, Gemini Omni and Gemini 3.5. Gemini Omni can create anything from any input, starting with video, and is a leap forward in world understanding, multimodality, and editing, while Gemini 3.5 Flash is the first in their latest family of models combining frontier intelligence with action.
With advancements to Google Antigravity, its agent-first development platform, they’ve moved beyond AI tools that just help write to agents that help act. Thanks to these agents, now anyone can be a builder.
Google is unlocking agents and agentic experiences across its products, like information agents in Search, Gemini Spark, and Daily Brief in the Gemini app, and the launch of Universal Card — a truly intelligent shopping cart.
In addition to that, Google is continuing to scale Gemini across all its products as well, from Google Photos to intelligent eyewear to Ask YouTube, creating new experiences and expanding to new form factors.
Read on to learn more about what’s new.
Key Takeaways
The Numbers Tell the Story
The scale of Google’s AI adoption is no longer theoretical. Token processing — the fundamental unit of AI workload — jumped from 480 trillion per month at last year’s I/O to more than 3.2 quadrillion today. That is a sevenfold increase in twelve months. More than 8.5 million developers are now building with Google’s models monthly. Thirteen Google products now have over a billion users each, five of which have surpassed three billion. AI Overviews alone has 2.5 billion monthly active users. The Gemini app has grown from 400 million to 900 million monthly active users in a single year, with daily requests increasing more than sevenfold over the same period.
These are not projections. They are the baseline from which Google is now building.
Two New Models, Two Different Bets
Gemini Omni is Google’s most ambitious model announcement. It can generate output in any modality — starting with video — from any input, combining Gemini’s language intelligence with Google’s generative media capabilities. The first release, Gemini Omni Flash, is now available on the Gemini app, Google Flow, and YouTube Shorts, with API access for developers and enterprises coming in the coming weeks.
Gemini 3.5 Flash is a different kind of announcement — less about novelty and more about the economics of AI at scale. It outperforms Gemini 3.1 Pro across almost all benchmarks while running 4x faster than comparable frontier models at less than half the price. Google’s own estimate: companies routing 80 percent of their workloads from other frontier models to 3.5 Flash could save more than $1 billion annually. For enterprises already burning through annual token budgets by May, that is a meaningful number. Gemini 3.5 Pro is expected to follow next month.
Also Read: How Network-as-a-Service Can Drive Sustainable IT
Antigravity 2.0: From Coding Tool to Agent Platform
Google’s internal development platform, Antigravity, has been reframed as the central infrastructure for the agentic era. Version 2.0 launches as a standalone desktop application — a control center for orchestrating cohorts of autonomous AI agents rather than a single coding assistant. Google says it has been using 3.5 Flash with Antigravity internally, and token processing across its AI developer tools grew from half a trillion to more than three trillion tokens per day between March and May. A version of Flash optimized specifically for Antigravity runs twelve times faster than other frontier models.
Gemini Spark: A Personal Agent That Runs While You Sleep
Gemini Spark is the consumer-facing materialization of the agentic vision. It is a personal AI agent built into the Gemini app that operates around the clock on dedicated virtual machines, handling long-horizon tasks in the background without requiring an open browser or active session. It integrates with Google’s own tools at launch and will extend to third-party tools through MCP in the coming weeks. On Android, task progress will be surfaced through a new interface called Android Halo, which will arrive later this year. Spark will also operate directly within Chrome as an agentic browser later this summer. Beta access begins for Google AI Ultra subscribers in the US next week.
Search Is Being Rebuilt for the Agentic Era
AI Mode, described as Google’s biggest Search upgrade ever, has already surpassed one billion monthly active users in its first year. Google is now taking information agents in Search further — personalized AI agents that operate continuously in the background, surfacing what a user needs at exactly the right moment. Generative UI capabilities will allow Search to build custom interactive experiences tailored to individual queries, available for everyone this summer at no cost. Persistent custom dashboards and trackers built within Search are also coming for Pro and Ultra subscribers, effectively turning Search into a platform for building lightweight personal applications.
Also Read: The AI Code No One Read Is Already in Production
The Infrastructure Behind the Ambition
All of this is running on Google’s eighth-generation TPUs, announced at Cloud Next. For the first time, Google has taken a dual-chip approach: TPU 8t for training, which delivers nearly three times the raw computing power of the previous generation and can distribute training across more than one million TPUs globally; and TPU 8i for inference, optimized for speed and energy efficiency, delivering up to two times better performance per watt. Capital expenditure is expected to reach between $180 and $190 billion this year — approximately six times the $31 billion Google was spending in 2022. The infrastructure investment is the least visible but arguably most consequential announcement of the day.
Transparency at Scale: SynthID Expands
As generative AI becomes more capable, Google is scaling its watermarking infrastructure alongside it. SynthID has now watermarked more than 100 billion images and videos, along with 60,000 years of audio assets. Content Credentials verification is being added across products, including Search and Chrome, to indicate whether content originated from a camera or AI and whether it has been edited with generative tools. Critically, OpenAI, Kakao, and ElevenLabs have all adopted SynthID — a cross-industry alignment on transparency infrastructure that represents one of the quieter but more significant developments of the conference.
Everything Else Worth Noting
Daily Brief, a new out-of-the-box agent in the Gemini app, synthesizes inbox, calendar, and task data into a personalized morning digest — prioritizing, organizing, and suggesting next steps rather than simply summarizing.
Google Pics, a new AI image creation and editing tool built on the Nano Banana model, treats every image element as an individual object rather than a flat file, enabling precise, element-level editing. It is rolling out to trusted testers now, with broader availability to Pro and Ultra Workspace subscribers later this summer.
Docs Live brings voice-first document creation to Google Docs — allowing users to verbally articulate their thoughts and let Gemini structure the output. It is rolling out to subscribers this summer, with voice capabilities coming to Gmail and Keep shortly after.
Ask YouTube reimagines content discovery by surfacing the specific segment of a video most relevant to a user’s question, rather than returning a list of videos for manual navigation. US rollout begins this summer.
Intelligent eyewear, previewed last year, is moving toward a fall launch. Audio glasses arrive first; display glasses — which surface navigation, translation, and contextual information in the lens — follow later.
Gemini for Science connects agentic platforms, including Antigravity to more than thirty major life science databases and tools, with the goal of accelerating research workflows at a scale that manual literature review cannot match.
Google I/O 2026 Day 2 coverage to follow.
