Understand the Core Announcements of Google I/O 2026 at a Glance

The 2026 Google I/O Developer Conference has concluded.

Looking back over the past six months, it seems that none of the excitement in the AI ​​world has anything to do with Google.

But anyone familiar with Google knows that it likes to save up its big moves and then unleash them all at once.

Finally, this year's I/O is here.

Today, I'm going to share with you what new things Google released this time.

Google has officially released two new model series: Gemini 3.5 Flash and Gemini Omni.

Gemini 3.5 Flash It became the absolute mainstay of this update, surpassing the previous generation flagship model Gemini 3.1 Pro in all benchmark tests across the three core dimensions of programming ability, multimodal understanding, and Agent tasks.

Its inference speed is 4 times faster than other cutting-edge large models in its class, and it achieved a high score of 83.6% on the Agent-specific benchmark MCP Atlas, outperforming GPT-5.5.

In terms of pricing, the Gemini 3.5 Flash costs $1.50 per million tokens to input and $9.00 per million tokens to output, which is three times more expensive than the Gemini 3.0 Flash, but 40% cheaper than the Gemini 3.1 Pro.

It seems that a price increase for tokens across the entire network is an inevitable trend.

As for the Gemini 3.5 Pro, they say, "Give us until next month to get it to you."

It is expected to be officially launched in June.

andGemini OmniThis has a fundamentally different positioning from Google's previous Veo series.

Veo's core capability focuses on generating videos from plain text; while Gemini Omni achieves true multimodal interaction—it supports any combination of input, including images, audio, video, and text, and has the dual capabilities of video generation and real-time editing.

You only need to upload a piece of original footage, and you can modify the people, backgrounds, and scenes in the video through natural language commands. It also supports the "partial preservation" function, which allows you to modify only a specified area while keeping the rest unchanged, which is a big plus.

Omni understands the physical laws of the real world, including gravity and fluid dynamics, which enables its generated video content to achieve a qualitative leap in logic and realism.

Google has explicitly stated that it will adopt a "cautious deployment" strategy for this feature. After all, technology that can modify people and content in videos without any barriers is undoubtedly a Pandora's box that needs to be treated with caution.

As a security safeguard, all videos generated by Omni are now automatically embedded with Google's SynthID digital watermark for content traceability.

As of today, the Gemini Omni Flash version has begun to be rolled out in stages through the official Gemini App, Google Flow, and YouTube Shorts platforms.

Google has updated several agents this time.

The first one is Gemini Spark, which you can think of as Google's version of OpenClaw.

Gemini SparkIt is a personal AI agent launched by Google, which runs on a virtual machine on Google Cloud and can operate 24 hours a day without you having to keep your computer on all the time.

It is powered by Gemini 3.5 Flash and Antigravity harness, enabling it to stably handle long-chain, highly complex asynchronous background tasks.

Furthermore, Spark is deeply integrated with Google's entire ecosystem of products, enabling it to seamlessly manage your digital affairs.

In a work setting, it can automatically organize your Google Docs documents, Gmail emails, and chat logs, extract the core information, and automatically draft emails strictly according to your preset writing style.

It also performs exceptionally well in everyday life. For example, when you're organizing a neighborhood party, Spark generates a real-time RSVP tracking sheet in Google Sheets and works in tandem with Gmail—the sheet automatically updates when a neighbor replies "I'll be there," and it automatically generates and sends personalized reminders to contacts who haven't replied.

So convenient, haha.

The second agent isAntigravity 2.0Version released.

Antigravity is a strategic product that Google invested $2.4 billion in, and Spark, mentioned above, is also built on the Antigravity platform. The product was first launched last November, and this I/O update brings its latest iteration.

FirstIt now has a brand new independent desktop application, which is different from the previous IDE plugin; it's now a true Agent working environment.

secondIt launched the Antigravity CLI, which essentially replaces the Gemini CLI.

According to an official Google announcement, the Gemini CLI and Gemini Code Assist IDE extensions will cease service for Pro/Ultra users after June 18, 2026.

If you are a developer using the Gemini CLI, please remember to migrate to the Antigravity CLI in advance.

thirdSimultaneously, the Antigravity SDK was launched, allowing developers to run the agent harness that Google uses in Antigravity directly on their own servers.

fourthIt integrates the Gemini audio model and is compatible with Android, Firebase, and AI Studio.

The engineer demonstrated on-site how Antigravity, in conjunction with Gemini 3.5 Flash, could build a working operating system from scratch entirely through natural language commands.

It can run command lines, run the Doom game, and play animations; it's very interesting.

It is worth noting that the Gemini 3.5 Flash has received exclusive deep optimizations on the Antigravity platform, and its inference speed is not the previously announced industry average of 4 times, but an astonishing 12 times.

Antigravity 2.0 is now available globally, and everyone can use it today.

AI Mode has surpassed 1 billion monthly active users, and its query volume has doubled every quarter since its launch.

At this conference, Google officially announced a complete upgrade of its underlying search model to Gemini 3.5.

Google says this is the biggest upgrade to the search box in 25 years.

Previously, you could only type questions, but now you can send in images, files, and videos, and the search will understand them across modalities.

Moreover, it uses AI to help you complete your questions and sort out the questions you really want to ask.

The Search feature now allows you to create an Agent. You input your needs, and it can automatically monitor the entire internet 24/7, proactively pushing suitable properties and sneaker collaborations to you.

You can also start multiple agents simultaneously in Search.

For example, writing code to generate custom gadgets, fitness tracking dashboards, and so on.

Google wants Search to become a personal assistant that proactively delivers information, rather than just a tool that answers your questions.

In his speech, Google CEO Sundar Pichai stated, "Search is increasingly becoming less of a one-off 'search and go' interaction and more of an ongoing, contextualized conversation."

It can build a fully customized interactive interface from scratch in real time for every question a user has.

This capability is underpinned by the Antigravity platform.

When a user initiates a search, the system automatically launches an independent containerized intelligent agent runtime environment, calls Gemini 3.5 Flash to write and execute code in real time, and directly embeds the final rendered interactive components into the search results page.

This feature will be available to all users for free in the summer of 2026.

Embedding dynamically generated UIs directly into search results is likely the most significant evolution in Google Search's product form since its inception in 1998.

Google, together with Samsung and Qualcomm, has officially unveiled the Android XR unified platform, with the first product to be launched – AI audio glasses – scheduled to hit the market in the fall of 2026.

These glasses feature a "head-up, hands-free interaction" experience; users simply need to say "..."Hey Google"You can wake up Gemini to query surrounding information, provide real-time walking navigation, send and receive messages and play music by voice, and even take photos from a first-person perspective with AI processing and analysis in real time."

To completely shed the stereotype of smart glasses as "technological experiments," Google partnered with two fashion brands, Gentle Monster and Warby Parker, to create two different designs: one avant-garde and trendy, and the other classic and everyday.

An advanced version equipped with a monocular microLED display will be launched later, and the entire product series will be compatible with both Android and iPhone platforms.

In terms of product strategy, Google's approach is very similar to that of the Meta Ray-Ban smart glasses, but the core difference lies in the complete Gemini ecosystem behind it, which theoretically provides more powerful Agent-level intelligent service capabilities.

This ultimate showdown in the field of smart wearables will officially kick off this fall.

Google has officially launched Universal Cart, a cross-platform AI shopping solution built on the new open-source Universal Business Protocol (UCP). Industry giants such as Amazon, Meta, Microsoft, and Salesforce have joined the protocol's ecosystem.

When you see a product you like on Google Search, Gmail, YouTube, or even any webpage, you can add it to this unified shopping cart with one click and complete the checkout in one stop.

It automatically monitors price fluctuations in the background, tracks historical price trends, sets restocking reminders, and can even intelligently identify product compatibility issues. In a live demonstration, when a user added an incompatible motherboard and CPU to their shopping cart, the system immediately issued a reminder and provided alternative solutions.

In closing.

At this conference, Google CEO Sundar Pichai unveiled a set of computing power data that has shaken the industry: Two years ago, Google's global big data model processed 9.7 trillion tokens per month; at last year's I/O conference, this number had soared to 480 trillion; and today, the number has reached 3.2 quadrillion—a sevenfold exponential growth in just one year.

"I never imagined I'd be speaking about 'quadri-trillions' on the I/O stage one day," Pichai remarked, highlighting the astonishing speed of development in the AI ​​industry.

The scale is there, the speed is there, and the direction is crystal clear. Frankly, Google announced far more products at this year's I/O than ever before, but all the announcements pointed to the same core strategy:Completely establish a closed-loop Agent ecosystem, allowing Gemini's capabilities to permeate every corner of users' digital lives.

When AI can truly handle your tasks 24/7 in the background, what are you prepared to let it do for you? This question is perhaps more worthy of our deepest reflection than "what Google released today".

Nebula Data, headquartered in Singapore, has branches in Jakarta, Guangzhou, Shanghai, and Hong Kong. The company independently developed Nebula Lab, a one-stop AI content generation and model aggregation platform, equipped with an enterprise-grade AI Agent, aggregating globally applicable large-scale models and industry-specific vertical models. Simultaneously, it launched the Nebula AIoT hardware ecosystem (including smart interactive terminals, IoT gateways, and other products), forming a full-link intelligent solution from cloud to edge to device. This provides integrated services to customers in e-commerce, manufacturing, retail, and other fields, from cloud computing power support and AI intelligent decision-making to terminal scenario implementation. Furthermore, it offers global AIDC (AI Intelligent Computing Center) + low-latency network services, empowering enterprises to embrace AI, connect to the physical world, and expand their global business through its technological foundation.