I/O '26 Recap: Everything You Need to Know

TL;DR: Google I/O 2026 announced Gemini 3.5 Flash/Pro, Gemini Omni, the Anti-Gravity agent platform, Gemini Spark personal AI, and comprehensive upgrades to Search, Shopping, and more—emphasizing full-stack AI innovation and scientific applications. ## AI Scale & Core Metrics Google AI is advancing on a full-stack basis. It now processes 3.2 quadrillion tokens per month (up from 97 trillion, a 7× increase). 13 products each have over 1 billion users, and 5 of those have over 3 billion. The Gemini model is the key driver of user growth. - **AI Overviews**: Over 2.5 billion monthly active users; AI Mode surpassed 1 billion MAU in one year. - **Gemini App**: Monthly active users grew from 400 million at last year's I/O to 900 million—more than double. ## New Products & Experience Upgrades ### Ask YouTube Users can directly ask YouTube questions like, "How do I teach a three-year-old to transition from a balance bike to a pedal bike?" The experience is completely reimagined. ### Docs Live Use voice to dictate ideas to Gemini. For example: "I just remembered there's an alumni lecture tomorrow… can you pull my résumé from Drive? … Extract details from it and put them at the top of the doc." The document is automatically generated and populated. ### Training Infrastructure Uses JAX and Pathways for seamless distributed training across sites, scaling to global TPU fleets to create the world's largest training clusters. ## Models & Speed ### Flash Model Demo Live creation of the Chrome Dino game: after the prompt, the response is generated in real time at nearly 1,500 tokens per second—far faster than the write request time. ### Gemini App Omni A new model that can create any output from any input (starting with video; future support for any input/output). It merges Gemini intelligence with generative media models for unprecedented multimodal understanding and editing. ### SynthID & Transparency SynthID has watermarked over 100 billion images and videos, and is now extended to Search and Chrome. OpenAI, I cacao, and Lemon Labs have also adopted SynthID 2. ## Gemini 3.5 Flash & Anti-Gravity ### 3.5 Flash Beats 3.1 Pro on all benchmarks, is faster, and sits in the top-right quadrant. Already used in the Anti-Gravity platform, speeding up builds by 12×. ### Anti-Gravity Agent Demo Task: Build a working operating system from scratch in 12 hours. 93 sub-agents worked in parallel, making over 15,000 model requests and processing 2.6 billion tokens. The initially empty project became a functional OS kernel. A live attempt to run *Doom* initially failed; an extra prompt made Anti-Gravity write over 100 lines of code to add video and keyboard drivers, and it ran successfully. ### 3.5 Pro In internal use with massive improvements; delivered to users next month. ## Gemini Spark: Personal AI Agent Spark runs 24/7 on a dedicated virtual machine in Google Cloud, using Gemini 3.5 and the Anti-Gravity framework to execute long-running background tasks. It integrates Google tools and will support third-party tools via MCP in the coming weeks. **Example scenario**: Organize a block party — Spark creates a real-time RSVP tracker (Google Sheets), records who brings what, and automatically emails neighbors who haven't signed up. Another task: Create a "Tasks to complete before the end of the school year" document, auto-formatting and linking follow-up actions. ## The Next Chapter of Search Google Search is AI search from end to end. Users can create and manage multiple AI agents within Search, set information agents to work in the background 24/7, and coordinate with Gemini Spark. **Example**: Interested in a biotech company, an agent connects to financial data in real time and sends synthetic updates when key changes occur to help understand the market. Search can build complete customized experiences: tools, trackers, dashboards. For instance, proactively build a weekend planner that securely connects Gmail, Photos, and Calendar, showing restaurant reservations and a map. ## Agent-Powered Commerce Users shop via Google more than a billion times every day. Three key building blocks: 1. Universal Commerce Protocol (UCP) 2. Agent Payment Protocol (AP2) 3. Universal Cart: A truly smart cart that works in the background. For example, while assembling a PC, the cart discovers the processor socket is incompatible with the motherboard and automatically suggests an alternative, preventing the issue. ## Gemini Experience: All-New Design Completely redesigned with Neural Expressive. New daily briefing — a personalized summary that combines inbox, calendar, and tasks, providing the most important info first thing every morning. ## Google Picks: Workspace Creative Tools Image creation and editing tools: hover to remove elements, resize objects, add/edit text, one-tap translation. ## Google Flow: Multi-Action Previously you could only execute one prompt at a time. Now the agent can take multiple actions simultaneously (analyze an image, brainstorm angles, generate 16 unique videos from a single image). ## Audio Glasses First-generation audio glasses launch this fall. Partnering with Samsung, Gentle Monster, Warby Parker, etc. Provides 24/7 help via Gemini, with voice delivered privately into the ear. Example: Say "order coffee" in a coffee shop, and Gemini automatically opens DoorDash, selects coffee, and confirms. ## Gemini for Science Brings together powerful AI tools to accelerate research. A new Labs prototype simplifies everyday scientific tasks: tracking new papers, translating research goals into code, generating new hypotheses. > Technology is the canvas for creativity, and AI will become the catalyst for a new golden age of scientific discovery. Source: https://www.youtube.com/watch?v=tfx2CjqtCUI