At Google I/O 2026, Google introduced an enormous wave of innovation throughout its know-how stack, marking a decisive shift into the agentic period of synthetic intelligence.
“Gemini 3.5 and Antigravity are unlocking a brand new world of brokers and agentic capabilities. We’ve been bringing brokers to builders and enterprises for some time. Now we’re tremendous targeted on bringing the facility of brokers, safely and securely, to shoppers in order that it really works for everybody,” Sundar Pichai, CEO of Google.
Pichai famous that this shift builds on a decade-long basis, monitoring again to when the corporate pivoted to be AI-first. Google continues to see AI as essentially the most profound option to advance its mission and enhance lives at scale, which is why it has taken a differentiated, full-stack strategy to AI innovation. This technique spans from customized silicon and safe infrastructure to world-class analysis, fashions, merchandise, and platforms that contact billions of individuals, finally enabling the corporate to iterate and innovate sooner in methods which are lighting up each a part of the organisation.
What’s unbelievable is how persons are utilizing AI, whether or not it’s college students prepping for remaining exams with the Gemini app, musicians and artists utilizing generative AI fashions like Lyria and Veo as a part of their artistic move, or builders coding and bringing their concepts to life.
AI momentum throughout the complete stack
To grasp the size at which persons are adopting AI, contemplate tokens, the basic models of information Google’s fashions course of. Two years in the past, Google processed 9.7 trillion tokens a month. Final 12 months, that grew to 480 trillion. Right this moment, that quantity has jumped 7x to over 3.2 quadrillion per thirty days. This scale highlights unbelievable demand throughout Google ecosystem:
Over 8.5 million builders construct experiences with Google fashions month-to-month
Google mannequin APIs course of roughly 19 billion tokens per minute
Over the previous 12 months, 375 Google Cloud clients every processed a couple of trillion tokens
Momentum with Google merchandise
Right this moment, Google has 13 merchandise with over a billion customers every, 5 of which cross the three billion person mark, pushed closely by Gemini integrations.
Search: AI Overviews reaches 2.5 billion month-to-month energetic customers. AI Mode has crossed 1 billion month-to-month energetic customers inside a single 12 months, shifting Search into ongoing, conversational interactions that join customers with the vastness of the net
Gemini App: Month-to-month energetic customers have greater than doubled in a 12 months to surpass 900 million, whereas each day requests grew over seven occasions. Options like Private Intelligence make responses deeply custom-made. Moreover, customers have generated over 50 billion photos to this point utilizing built-in Nano Banana fashions, proving to be a breakout success for latent creativity worldwide
Pure, conversational AI in merchandise
Google is bringing pure conversations with Gemini immediately inside on a regular basis merchandise. Not too long ago, Maps obtained its greatest improve in a decade with Ask Maps, permitting customers to ask complicated, long-form questions. Now, Google is increasing these intuitive conversational capabilities elsewhere.
Ask YouTube
Ask YouTube totally reimagines video discovery, making deep video data digestible and simple to navigate. The function surfaces movies that finest match customers’ nuanced pursuits and jumps straight to essentially the most related section. Testing begins now, with a broad U.S. rollout scheduled for this summer season.
Voice-powered Docs Dwell
Enabled by fast developments in Google audio fashions, Docs Dwell permits customers to verbally “mind dump” no matter is on their thoughts to immediately generate paperwork. Transferring past exact typed prompts, customers can create and edit recordsdata totally with their voice. Docs Dwell rolls out for subscribers this summer season, with voice options coming to Gmail and Preserve quickly after.
Infrastructure supporting innovation at scale
Supporting this world scale requires large infrastructure investments. Transferring from $31 billion in annual capex in 2022, Google expects to spend roughly $190 billion this 12 months, targeted closely on customized silicon.
A decade after debuting Google’s first TPU, the corporate has launched its eighth era of customized silicon: TPU 8t and 8i, using a dual-chip strategy optimized individually for coaching and inference:
TPU 8t: Optimized for large-scale pretraining, delivering thrice the uncooked computing energy of its predecessor. Distributed throughout a number of information facilities by way of JAX and Pathways, it scales throughout greater than 1 million TPUs globally to type the most important coaching cluster on the planet.
TPU 8i: Tailor-made particularly for high-speed inference, dramatically decreasing latency the place it issues most.
Each chips ship as much as two occasions higher performance-per-watt, making certain sustainable scaling.
Gemini 3.5 Flash
Right this moment, Google is introducing Gemini 3.5 Flash, the primary in a sequence of fashions combining frontier intelligence with fast execution.
Efficiency & Pace: Outperforming 3.1 Professional throughout key benchmarks, 3.5 Flash exhibits a rare soar in coding and real-world economically precious duties (GDPVal). It operates 4 occasions sooner than different frontier fashions.
Inner Scale: Paired with Google’s agent-first improvement platform, Antigravity, inside utilization throughout the corporate’s developer instruments has grown to greater than three trillion tokens a day.
Value Effectivity: Flash delivers frontier capabilities at lower than half the value of comparable fashions. If high enterprises shifted 80% of their workloads to three.5 Flash, they’d save over $1 billion yearly.
Gemini 3.5 Flash is on the market at this time throughout Google’s merchandise and APIs. Gemini 3.5 Professional is getting used internally and arrives subsequent month.
Antigravity 2.0
Google can be bringing 3.5 Flash to builders in Antigravity, remodeling the ecosystem right into a platform to handle cohorts of autonomous AI brokers. Antigravity 2.0, a standalone desktop software, serves as a central dwelling for agent interplay. It makes use of a model of Flash optimized to run 12x sooner than different frontier fashions, obtainable for customers to expertise beginning at this time.
Brokers
Google is targeted on delivering safe, consumer-ready AI brokers that deal with complicated, multi-step workflows.
Gemini Spark: Working 24/7 on Google Cloud digital machines, Spark is a private AI agent constructed on Gemini 3.5 and the Antigravity harness. It executes long-horizon background duties, connects with instruments by way of MCP, and integrates with the Gemini app. It expands to e mail, chat, and Chrome this summer season, with real-time standing monitoring arriving by way of Android Halo later this 12 months. U.S. Beta testing begins subsequent week for Google AI Extremely subscribers.
Agentic Search: Google is introducing Data Brokers in Search to constantly collect insights and carry out actions 24/7, rolling out this summer season for AI Professional and Extremely subscribers. Search can even function Generative UI Capabilities to construct free, custom-made layouts and interactive visuals for complicated queries, alongside Persistent Customized Dashboards to trace long-running tasks.
Extra from Google agentic Gemini period
Google previewed further agentic ecosystem toolsets and collaborative workflows:
Every day Temporary: A customized, extremely concise morning digest for the Gemini app that prioritizes, organizes, and suggests subsequent actions immediately from a person’s inbox, calendar, and duties.
Google Circulation: A collaborative workspace agent that plans, causes, and assists with challenge brainstorming. It introduces “vibe coding” options, permitting customers to quickly generate artistic belongings like video results and layered textual content.
Google Pics: A canvas modifying device constructed on the Nano Banana mannequin that isolates design parts as distinctive objects, letting customers regulate particular particulars dynamically. Out there now to trusted testers.
Clever Eyewear: Gemini-powered audio glasses (providing spoken ear help, launching this fall) and show glasses designed for hands-free contextual data monitoring.
Gemini for Science: An experimental Analysis Lab structure linking platforms like Antigravity to over 30 life science databases, leveraging Deep Suppose and Deep Analysis to speed up discoveries.
