Gemini 3.5 Unveiled: A Complete Roundup of Google I/O 2026's Agentic Era Announcements

Google I/O 2026 opened with a bold declaration: we’ve entered the agentic era, where AI agents don’t just respond to commands—they proactively execute complex tasks, understand the physical world, and integrate seamlessly into daily life. This year’s keynote was packed with game-changing announcements, from next-gen Gemini models to agent-first development platforms and AI-powered hardware. Here’s the complete roundup of every major reveal.

Gemini 3.5 Flash: The High-Speed AI Workhorse

The star of the show, Gemini 3.5 Flash, is now the default model across the Gemini app, Search AI Mode, and Gemini API. Its sibling, Gemini 3.5 Pro, is in closed beta and set to launch next month. Performance-wise, it outpaces Gemini 3.1 Pro across key benchmarks:

1. Terminal-Bench 2.1 coding: 76.2% accuracy (up from 70.3%)
2. GDPval-AA real-world tasks: 1656 Elo score (vs. 1314)
3. MCP Atlas tool calling: 83.6% success rate (up from 78.2%)
4. CharXiv multimodal reasoning: 84.2% accuracy

Speed is another standout: it generates output at 289 tokens per second—four times faster than competing leading models. With Antigravity optimization, that speed jumps to 12x. Pricing is equally aggressive: $1.50 per million input tokens, $9.00 per million output tokens, and just $0.15 per million cached input tokens, with a 1M token context window. This is over 50% cheaper than comparable models, and top Cloud customers have already switched 80% of their advanced workloads, saving over $10 billion annually.

Adoption is skyrocketing: Google’s internal developers now use 3 trillion tokens daily, a 6x increase in two months, while over 375 external customers each process more than 1 trillion tokens monthly.

Gemini Omni: The Multimodal World Model

Google introduced Gemini Omni, a unified multimodal model that merges Gemini’s reasoning power with the generative capabilities of Veo, Nano Banana, and Genie. It doesn’t just process text and images—it understands real-world physics, spatial relationships, and causal chains, enabling it to generate any modality from any input.

Key use cases include:

1. Physics simulation: Generating scientifically accurate clay animations of protein folding
2. Conversational video editing: Modifying selfie video styles, adding elements, or switching perspectives while preserving original actions

Gemini Omni Flash is now live in the Gemini app, Google Flow, and YouTube Shorts (free for Shorts users). Access in the Gemini app and Flow requires an AI Plus, Pro, or Ultra subscription, with Omni Pro coming soon.

Antigravity 2.0: Building Agents at Scale

Antigravity has evolved from an IDE extension to a full agent-first development platform, available in four forms: a standalone desktop app (supporting macOS/Linux/Windows), CLI, SDK, and Managed Agents within the Gemini API.

The desktop app serves as an agent orchestration tool, supporting multi-agent parallel workflows with new features like scheduled tasks, JSON Hooks, project-level permissions, and /goal slash commands. In a demo, developers used Gemini 3.5 Flash and subagents to build a functional operating system running Doom in just 12 hours—using 93 subagents, 15,000 model requests, 2.6 billion tokens, and costing less than $1000.

The CLI, rewritten in Go, replaces the Gemini CLI, which will shut down along with the Code Assist IDE plugin on June 18, requiring user migration. The SDK lets teams define and deploy custom agents to their own infrastructure, while Managed Agents let users launch isolated Linux sandbox agents with a single API call, retaining state and supporting markdown-extended skills.

AI Studio has also been updated, integrating Antigravity’s coding agent and adding a mobile app that generates native Android/Kotlin/Jetpack Compose apps with an in-browser simulator and Play Console internal test track support. The first two Cloud deployments are free.

Gemini Spark: Your Personal AI Agent

Powered by Gemini 3.5 Flash and the Antigravity Harness, Gemini Spark is a private AI agent running on dedicated Google Cloud VMs, available 24/7 via the Gemini app, email, and chat.

Key capabilities include cross-workspace task execution across Gmail, Drive, Sheets, Slides, and Calendar; custom skill creation; voice input that splits into multiple tasks; and pause-and-approval for sensitive actions. On Mac, users can select a file and voice-generate an email with tables, with the agent automatically correcting ambiguous instructions.

Timeline and pricing: Trusted testers get access this week, with a beta rolling out to U.S. AI Ultra subscribers next week. A new $100/month Ultra plan launches, while the previous top-tier $250/month package drops to $200/month. It will arrive in Chrome this summer, Android Halo later this year, and soon integrate with MCP, with Instacart as a launch partner.

Google Search: Reinvented for the Agentic Age

Search’s AI Mode has been upgraded to Gemini 3.5 Flash, boasting over 1 billion monthly active users and query volumes doubling every quarter—with last quarter’s total Search queries hitting an all-time high, and AI Overviews reaching 2.5 billion monthly users.

The biggest update in 25 years is the new search box, supporting multimodal input (text, images, files, videos, Chrome tabs) with real-time AI suggestions, launching globally today. AI Overviews and AI Mode now merge, letting users slide from the main results page’s AI answer into AI Mode for contextually coherent follow-ups—live today on desktop and mobile.

Search Agents will let users create 24/7 background information agents to handle complex tasks like tracking stocks under specific conditions or finding housing, with integration with Gemini Spark coming this summer. Generative UI, built with Antigravity’s coding capabilities, will create custom interactive interfaces for search questions—free to use this summer, with subscription users able to build their own.

Ecommerce Reimagined: Three Game-Changing Tools

Google unveiled three ecommerce innovations to streamline agent-driven transactions:

1. Universal Commerce Protocol (UCP): An open-source co-developed with Amazon and Meta, defining a universal language for agents and ecommerce systems across the entire customer journey. It expands to hotels, local delivery, and YouTube, launching soon in Canada, Australia, and the UK.
2. Agent Payments Protocol (AP2): Lets users set purchase boundaries, enabling automatic purchases when conditions are met. It uses tamper-proof digital authorization letters with privacy encryption, and shares return records between parties—launched first with Gemini Spark.
3. Universal Cart: A cross-platform cart across Search, Gemini, YouTube, and Gmail that automatically tracks price drops, stock levels, discounts, and product compatibility, matching Google Wallet credit card offers. It launches in the U.S. this summer on Search and Gemini app, followed by YouTube and Gmail.

Gemini App & Creative Tools: Redesigned for Creativity & Productivity

The Gemini app gets a full redesign with Neural Expressive, featuring fluid animations and custom content generation, plus dialect support—live today on Android, iOS, and web globally.

New tools include:

1. Daily Brief: Automatically summarizes and organizes information from inboxes and other sources with actionable next steps—live today for U.S. AI Plus/Pro/Ultra subscribers.
2. Gemini Omni in App: Paid subscribers can now use Omni for video creation and editing in the app, available today.
3. Google Pics: A Workspace tool for AI image creation and editing with SynthID watermarking, launching this summer.
4. Stitch: A UI design tool with real-time voice collaboration, supporting code export and one-click website publishing—live globally today.
5. Google Flow: Now supports Gemini Omni, generating 16 different camera-angle videos from a single image, plus new Flow Agent, Flow Tools, and Flow Music—all live today.
6. Docs Live: Lets users generate and edit documents via voice input, automatically pulling materials from Drive and Gmail—launching this summer for Pro/Ultra subscribers, with sync to Gmail and Keep.

Smart Glasses: AI at Your Fingertips (and Eyes)

Google partnered with Samsung for hardware and Gentle Monster/Warby Parker for design to launch two smart glass models, compatible with Android and iOS:

1. Audio Glasses: Launching this fall, featuring bone-conduction Gemini voice conversations, hands-free navigation, and app action delegation (like ordering coffee).
2. Display Glasses: Currently in testing, with a small in-lens screen for viewing ride-sharing info, real-time translation, and custom interfaces. Developers are building experiences, with more details coming later this year.

DeepMind: Advancing Science & AI Safety

DeepMind showcased breakthroughs in scientific research and AI safety:

1. Gemini for Science: Includes three experimental tools—Literature Insights, Computational Discovery, Hypothesis Generation—paired with the Co-Scientist multi-agent research partner.
2. AlphaEarth Foundations: A planet-scale digital twin for simulating dynamic systems like forest degradation.
3. WeatherNext: Accurately predicted Hurricane Melissa’s path 3 days early during the 2025 hurricane season, outperforming traditional models, and is now part of the U.S. National Hurricane Center’s regular toolkit.
4. Isomorphic Labs: Accelerating drug development via molecular interaction modeling, with multiple projects in preclinical stages.
5. Safety Tools: SynthID has watermarked 100 billion images, videos, and 60,000 years of audio, with OpenAI adopting the technology. Content Credentials Verification expands to Search and Chrome, plus a new AI Content Detection API. The CodeMender safety agent’s API is now open for testing.

From foundational model upgrades to agent-first development tools and AI-integrated hardware, Google I/O 2026 paints a clear picture of the agentic era’s potential. Every announcement ties back to a single vision: making AI agents accessible, powerful, and safe for everyone. As these tools roll out in the coming months, we’re not just seeing incremental updates—we’re witnessing a paradigm shift in how we interact with technology.