Google I/O 2026 Kicks Off: A Flood of Announcements in Just One Event

In the early hours of May 20, 2026 (Japan time), Google held its annual developer conference, “Google I/O 2026.” Over the past six months, companies like OpenAI and Anthropic have dominated the generative AI space with frequent updates, but Google stuck to its strategy of stockpiling information and releasing everything at once during I/O. True to tradition, this year’s event unveiled a massive array of updates spanning AI models, ecosystems, agents, and infrastructure.

This article highlights the key points you need to know.

Advances in Core AI Models

Gemini 3.5 Flash: A Compact Model Outperforming Previous Generation Flagships

One of the main highlights this year is the official unveiling of Gemini 3.5 Flash. Traditionally, the Flash series has been positioned as a sub-flagship model known for being lightweight, fast, and affordable. However, following the industry trend of compact models surpassing the performance of previous-generation flagship models, Gemini 3.5 Flash has significantly outperformed its predecessor, the 3.1 Pro, across multiple benchmarks.

Specifically, it scored 76.2% on the Terminal-Bench 2.1 test for coding capabilities, compared to 70.3% for the 3.1 Pro. On the GDPval-AA benchmark, which measures real-world economic value, the 3.5 Flash achieved an Elo score of 1656, surpassing the 3.1 Pro by over 300 points (1314 Elo). Its agent functionalities and tool integration capabilities also far exceed those of the previous generation.

However, it isn’t superior across all aspects. On the “Humanity’s Last Exam,” which tests global knowledge and abstract reasoning, it scored 40.2%, trailing the 3.1 Pro’s 44.4%. Similarly, it recorded 72.1% on the ARC-AGI-2 benchmark, compared to the 3.1 Pro’s 77.1%. This suggests that the model emphasizes practical capabilities over abstract reasoning and knowledge.

The output speed is reportedly four times faster than other state-of-the-art models. Pricing stands at $1.50 per million input tokens and $9.00 for output tokens, making it 40% cheaper than the 3.1 Pro. It supports a context window of 1 million tokens and has a knowledge cutoff date of January 2025. Starting today, it is the default model for the Gemini app and search AI mode, and its API is also available. Additionally, Gemini 3.5 Pro is slated for release next month.

Gemini Omni Flash: A New Force in Multimodal Generative AI

Another significant announcement is the debut of Gemini Omni Flash, a multimodal model capable of generating content based on any input. Positioned as a competitor to Seedance 2.0 in the video generation domain, it also features the ability to modify specific parts of a video while keeping other sections intact. However, some have noted room for improvement in the overall user experience. The Omni Pro model is expected to be released at a later date.

Overhauling the Gemini Ecosystem

Major Redesign of the Gemini App

The Gemini app has adopted a new “Neural Expressive” design language, featuring an overhauled interface with a distinctive blue gradient background. Functional entry points have been consolidated under a ”+” symbol. A new feature allows users to adjust the depth of AI reasoning, and usage limits have been introduced. The redesigned app is now available globally.

Integration of AI Features into Major Google Products

Google Maps now includes an “Ask Maps” feature, enabling users to ask complex questions about directions or locations in natural language. YouTube has introduced “Ask YouTube,” which answers direct questions about video content and pinpoints relevant scenes.

Google Docs now supports real-time voice-based document creation, allowing users to generate and organize documents entirely through voice interaction. This feature will be available to Pro and Ultra subscribers this summer.

Additionally, a personalized morning summary feature called “Daily Brief” has been launched and is now available to subscribers in the United States.

Significant Updates to NotebookLM

NotebookLM has received a comprehensive update, including features like video summary generation, creation of infographics in multiple styles, and syncing learning progress across devices. Integration with the Gemini app has been enhanced, enabling EPUB file uploads and PPTX exports. It can now also connect with Google Classroom, making it more useful in educational settings.

Agent Systems as the Core Theme of I/O 2026

Antigravity 2.0: A Major Leap in Development Platforms

At the heart of this year’s announcements is the expansion of agent systems. The Antigravity 2.0 development platform now includes a standalone desktop application and a command-line interface (CLI), set to replace the old Gemini CLI on June 18, 2026. An SDK has also been released, allowing developers to deploy their own solutions.

With native integration of voice capabilities and optimization for Gemini 3.5 Flash, processing speeds have increased by up to 12 times. In a live demo, 93 parallel sub-agents operated continuously for 12 hours, handling 15,000 model requests and processing 2.6 billion tokens while building an operating system from scratch. The total cost was reportedly under $1,000.

Gemini Spark: A Personal AI Agent Operating 24/7

Gemini Spark is a personal AI agent that runs on virtual machines exclusive to Google Cloud, capable of processing tasks in the background 24/7. It integrates with Google services to automate time-consuming tasks like summarizing emails and preparing events.

The subscription model has been adjusted, with the Ultra plan starting at $100 per month. Beta testing for Ultra subscribers in the U.S. begins next week.

Android Halo: A Dedicated Agent Layer for Android

For Android devices, Google unveiled Android Halo, a dedicated background layer for agents. It displays the agent’s progress in real-time on the status bar. This is Android’s first UI layer specifically for agents and is not intended for apps. It is scheduled for release later this year.

New Frontiers and Infrastructure Development

Expansion of Generative Creative Tools

The image creation tool “Google Pics” was announced, supporting single-element editing and automatically adding AI watermarks. The UI design tool “Stitch” now features real-time voice collaboration and one-click publishing.

Google Flow, now integrated with Gemini Omni, supports bulk generation of multi-angle videos and scene-wide modifications. A new “Flow Music” tool can create complete compositions from audio fragments, aiming to cover the entire creative workflow ecosystem.

SynthID, Google’s AI watermarking technology, now allows right-click searches in Chrome. This initiative, supported by major players like OpenAI, aims to standardize AI-generated content.

Major Upgrade to Search AI

Google’s search AI mode has surpassed 1 billion monthly active users (MAUs), with query volumes doubling each quarter. The underlying platform has been upgraded to Gemini 3.5, marking the most significant overhaul of the search box in 25 years. It now supports multimodal input and automatic question completion, with integrated AI summaries and conversation modes.

A new background search agent enables continuous information monitoring, and “Agentic Coding” was introduced to provide real-time interactive visualizations. This feature will be freely available to all users this summer and is described as the most significant evolution of Google Search since 1998.

Implementation of Agent-Based E-Commerce Infrastructure

The universal e-commerce protocol “UCP” was announced, with participation from key players like Amazon and Microsoft. It will initially launch in the U.S. before expanding internationally. A new payment approval protocol, “AP2,” introduces consumer protection safeguards.

A cross-platform smart cart called “Universal Cart” was also introduced, allowing seamless cart additions across scenarios. It supports automatic discount searches and compatibility checks and is set to launch in the U.S. this summer.

Other Major Announcements

Google’s first audio glasses for Android XR will be released this fall, supporting AI interactions and cross-device operations.

The 8th generation TPU introduces separate chips for training and inference, with training chips offering three times the computational power of the previous generation. Inference chips achieve a generation speed of approximately 1,500 tokens per second.

“Gemini for Science,” a toolkit for scientific research, was also announced. It includes a more precise AI weather forecasting tool called “Weather Next” and advancements in AI-driven drug discovery, with several projects entering preclinical stages. An open test for a code vulnerability auto-repair tool named “Code Mender” has also begun.

Conclusion: Google’s AI Strategy Becomes Clearer

Google I/O 2026 showcases the company’s commitment to not just offering high-performance models but also building an entire ecosystem centered around agent systems. Features like 24/7 personal agents and Android Halo could fundamentally change how users interact with technology.

The trend of compact models like Gemini 3.5 Flash outperforming previous-generation flagships points to a significant step toward democratizing AI and reducing usage costs. The milestone of 1 billion monthly active users for search AI further underscores the mainstream adoption of AI technologies.

With the upcoming releases of Gemini 3.5 Pro, Omni Pro, Android XR glasses, and more, Google’s momentum is only expected to accelerate, setting the stage for the future of the tech industry.