谷歌I/O大会All in AI：Gemini模型上新，搜索框25年来最大改版，AI眼镜回归，Spark智能体首秀

Original Author: Li Dan

Original Source: Wall Street News

On Tuesday, May 19, US Eastern Time, Google, a subsidiary of Alphabet, released a flurry of AI updates at its annual developer conference, Google I/O 2026. From redesigning Google Search to launching the Gemini Omni and 3.5 Flash models, as well as the Gemini Spark agent, to the first public unveiling of its next-generation AI glasses, Google is attempting to embed AI into the entire internet entry point ecosystem—search, Android, office productivity, video, and next-generation hardware.

Alphabet and Google CEO Sundar Pichai revealed during the keynote that the Gemini application (app) has doubled its monthly active users to over 900 million in a year; the "AI Overviews" feature in Google Search now has over 2.5 billion monthly active users; and the "AI Mode" has surpassed 1 billion users.

Pichai emphasized that AI is no longer just a chatbot but the "next computing platform." He stated at the conference: "AI is illuminating every part of the company."

From the Gemini Omni model and Gemini 3.5 Flash showcased at this conference, to the Gemini Spark agent capable of continuously executing tasks, and the restructuring of its AI subscription system, Google is attempting to upgrade Gemini from a chatbot into an "AI operating system" covering search, office, development, video, and the mobile ecosystem.

This conference is also seen externally as a comprehensive counterattack by Google against competitors like OpenAI and Microsoft. Over the past year, as ChatGPT has changed user search and content consumption habits, Google's traditional search business has faced its biggest challenge since inception. The I/O conference clearly demonstrates its latest strategy: leveraging the world's largest internet traffic gateway to embed AI "natively" into the entire Google ecosystem.

The Biggest Redesign of Search in 25 Years: The Search Box Becomes an AI Entry Point

One of the most core changes at this year's I/O is Google's complete AI transformation of its search business.

According to reports, Google is redesigning its iconic search box to better handle the "longer, more complex questions" users pose to AI chatbots.

The new search box will:

• Support more natural language input;
• Make it easier to upload images and files;
• Automatically assist users in refining their searches;
• Support AI agents for continuous task execution.

Nick Fox, head of Google's Knowledge and Information Platform, said: "Users should be able to type any question that's on their mind directly into the search box."

Google also plans to integrate Agent functions into Search to help users: track topics of interest; book services; manage health information; and create personalized task dashboards. Some of these features will only be available to paid subscribers.

Industry observers believe this signifies Google Search is transitioning from a traditional "link distributor" to an AI task entry point.

Google's Most Powerful Multimodal Model, Gemini Omni, Makes Its Debut

One of the most significant updates at this year's conference is the launch of Google's new Gemini Omni model.

Multiple technology media outlets report that Gemini Omni is positioned as Google's most advanced multimodal AI system, capable of simultaneously processing text, images, video, voice, and real-time interactive tasks.

Compared to previous Gemini series, the biggest changes in Omni are:

• Greater emphasis on real-time interaction;
• Better suited for Agent scenarios;
• Ability to perform continuous tasks across applications and devices.

Industry experts believe Google is using this to directly compete with OpenAI's GPT-5 Agent system and Microsoft's evolving Copilot ecosystem.

Google executives repeatedly emphasized at the conference that "AI is not just about answering questions but about actually getting work done." This shift indicates Google's AI strategy has moved from "chat assistant" to "task execution platform."

Gemini 3.5 Flash Focuses on Speed and Low Cost; 3.5 Pro Arriving in June

In addition to the flagship model, Google also introduced Gemini 3.5 Flash.

According to information from the I/O conference and Google's developer pages, the Flash series is positioned for low latency, high response speed, and lower inference costs, targeting: developers; mobile devices; real-time interactions; and AI programming/automation scenarios.

As a new lightweight member of the Gemini model family, Gemini 3.5 Flash possesses cutting-edge technical capabilities, priced at half the cost of comparable frontier models, and in some cases as low as one-third. Google announced that Gemini 3.5 Flash will become the global default model for the Gemini App and "AI Mode" within Google Search.

Google stated it has strengthened the cybersecurity of Gemini 3.5 Flash, making it "less likely to generate harmful content and also less likely to incorrectly reject safe queries."

Google also revealed that its more powerful version, Gemini 3.5 Pro, is currently being used internally but will be released to a wider user base next month, in June.

Media outlets pointed out that Google is clearly aware that the AI competition is no longer just about "who is smartest," but about who is faster, cheaper, and better suited for large-scale deployment.

In the age of Agents and real-time AI, the importance of latency and cost is rapidly increasing.

Reports suggest that Gemini 3.5 Flash will become one of the foundational models for Google's new generation of AI subscription system.

Gemini Spark: Google Officially Bets on "AI Agents"

Compared to model updates, Gemini Spark attracted more market attention.

Google defines it as a "continuously running AI agent" capable of performing tasks in the background over the long term. Multiple media outlets have called this one of the most strategically significant announcements of this year's I/O.

According to Google's demonstrations, Gemini Spark will be able to:

• Automatically organize emails;
• Monitor user schedules;
• Track web page information;
• Manage documents;
• Complete cross-application workflows;
• Even make purchases and bookings on behalf of the user.

Reports indicate Google is promoting the concept of "always-on AI," where AI no longer waits for user commands but proactively handles tasks.

Google Vice President Josh Woodward stated: "Spark represents a major shift for Gemini." Industry experts believe this signifies the AI industry is moving from the "chatbot era" into the "agent era."

Previously, whether ChatGPT or Gemini, they were essentially "conversational tools"; Spark, however, is closer to a true digital assistant.

Google clearly hopes to leverage the ecosystem of Gmail, Chrome, YouTube, Docs, and Android to turn Gemini into an AI execution entry point at the internet level.

"We Are Entering the Age of Search Agents"

Google executives stated at the I/O conference: "We are entering the age of search agents." Google Search has achieved a performance leap with the support of the Gemini 3.5 model, enabling it to perform complex tasks like creating travel itineraries through search.

Robby Stein, Vice President of Product for Google Search, demonstrated the agentic capabilities of Gemini. He said: "We firmly believe that the best search experience should be one that truly works for you."

With the new "AI Overviews" feature in Google Search, users will experience highly interactive searches—for instance, delving into the mysteries of black holes. Depending on the search topic, the feature can even automatically generate corresponding animations and customized user interfaces.

Google Search now possesses the ability to perform tasks at scale. It can automatically build mini-apps for users, helping to efficiently handle large-scale tasks. A typical use case is when a user wants inspiration for weekend family entertainment; Google Search can create a detailed weekend activity plan from scratch.

Google and Samsung's AI Glasses Launching This Fall

Hardware was also a key theme at this year's I/O.

Google publicly showcased its next-generation AI smart glasses for the first time, marking its return to the smart glasses market. Google announced partnerships with Samsung and eyewear manufacturers Gentle Monster and Warby Parker to launch smart glasses with the built-in Gemini assistant. Google stated these glasses will be compatible with Android and iOS devices and are expected to launch later this year.

Shahram Izadi, Head of Product and Platform for Android XR, said: "These are just the first two designs in a much larger lineup set to launch this fall." He pointed out that the glasses are designed to deliver information "privately to the user's ear, rather than through a screen display."

Samsung stated the device will combine the Android XR platform with Gemini AI capabilities to create an "intelligent wearable computing experience."

Demonstrations at the I/O conference showed these AI glasses can: translate in real-time; recognize the surrounding environment; provide navigation; offer voice interaction; and call upon Gemini agents.

Industry experts believe Google is attempting to replicate the ecosystem advantages Android had in the smartphone era, combining AI agents with AR devices to preemptively stake a claim in the next computing platform.

Revamping AI Subscriptions: Launching a $100/Month Developer Plan

Beyond models, another significant change at the conference was Google's redesign of its AI pricing structure.

Google announced a new AI subscription plan for developers, priced at $100 per month. The plan will offer:

• Higher Gemini API call quotas;
• Advanced AI Studio features;
• Gemini Code Assist;
• Google Cloud credits;
• Higher-tier Agent access permissions.

Analysts believe this indicates Google is fully committing to developer monetization.

Previously, OpenAI and Anthropic had established first-mover advantages in the enterprise and developer markets. While Google possesses powerful model capabilities, its developer ecosystem progress has been relatively slow.

Now, Google clearly hopes to re-attract developers by offering lower prices and deeper tool integration.

AI Ultra Monthly Subscription Reduced from $250 to $200

Meanwhile, Google also adjusted its flagship AI subscription plan, AI Ultra.

Previously priced at $250 per month, Google announced a price reduction to $200.

AI Ultra currently offers: access to the highest-tier Gemini models; Deep Think reasoning capabilities; advanced Agent functions; AI video and image generation; AI programming capabilities; advanced NotebookLM features; and higher API and context limits.

The market views this price adjustment as a sign that the AI subscription war is rapidly intensifying.

Over the past year, OpenAI launched ChatGPT Pro; Anthropic strengthened Claude Max; Microsoft expanded Copilot enterprise subscriptions; and xAI also promoted Grok premium memberships. Google's reduction of Ultra's price is aimed at expanding its user base and signals that the competition among large models is shifting from a "technology war" to a "price war."