Google I/O 2026: Gemini 3.5 Flash, Gemini Omni, and Antigravity 2.0 — Everything Announced Today

Google I/O 2026 kicked off today, May 19, in Mountain View with Sundar Pichai and Demis Hassabis on stage. The two-day event runs through May 20 and is available for free streaming on Google's official YouTube channel and at io.google.

As expected, AI dominated the announcements. Here is everything that matters for developers and technical users.

Gemini 3.5 Flash: The New Reference Model for Developers

Google officially announced Gemini 3.5 Flash, the model that has been leaking through Google's infrastructure for weeks as "Gemini 3 Fast". According to data presented at the keynote, Gemini 3.5 Flash combines frontier-level intelligence with the ability to execute actions — not just generate text, but interact with services and complete complex multi-step tasks.

The model is available starting today in Google AI Studio and the Vertex AI API. The proposition is clear: higher capability than Gemini 3.1 Flash at a competitive cost, with support for 1 million token contexts.

Antigravity — Google's agentic development platform — now uses Gemini 3.5 Flash as its base model. The on-stage demos showed the model building complete systems from scratch, including a demo where Antigravity generated a functional operating system capable of running Doom.

Gemini Omni: Any Input, Any Output

The most ambitious announcement of the keynote is Gemini Omni, presented by Demis Hassabis as the next step in Gemini's architecture. The proposition is a unified architecture that integrates text, image, audio, and video in both input and output — the model can generate any type of content from any type of input.

The first available version is Gemini Omni Flash, which is live starting today. Highlighted capabilities include conversational video editing — you describe the changes you want applied to a clip and the model executes them — and visual content generation with physics understanding for more realistic interactions.

Gemini Omni also integrates content credential understanding: it can identify whether an image or video was AI-generated. Google announced that OpenAI, NVIDIA, and ElevenLabs will adopt SynthID — Google DeepMind's content watermarking technology — to make AI-generated content identifiable.

Antigravity 2.0: Google's Agentic Development Platform

Antigravity 2.0 was presented as Google's agentic development platform, described as "unabashedly agent-first" — designed from the ground up for building agents, not just traditional applications.

Concrete announcements available starting today:

Antigravity CLI: command-line interface for integrating agentic capabilities into existing development workflows
Antigravity SDK: development kit for building agents with access to Google tools
Native Voice Support: native voice support in agents built with Antigravity
Antigravity 2.0 Desktop App: desktop application with agent mode capable of working with local folders, connectors, and custom skills

All available globally starting today. Integration with Firebase and Google AI Studio lets developers deploy agents directly in their existing projects.

Android XR: Gemini-Powered Smart Glasses

Google confirmed the Android XR glasses presentation at today's keynote — something that had been teased for months. Hardware partners include XREAL, Warby Parker, and Gentle Monster, pointing to products across different price ranges.

The glasses integrate Gemini directly, with real-time information overlay capabilities on the physical environment. Google is competing in this segment directly with Meta Ray-Ban glasses and Apple Vision Pro.

Consumer availability dates have not been confirmed at the time of publishing this article.

Googlebook: The New Gemini-Native Device Category

Google provided more details on Googlebook, the new category of laptops with Gemini integrated at the system level that was announced at the Android Show last week. The first Googlebooks will arrive in fall 2026 from manufacturers including Acer, ASUS, and Lenovo.

Confirmed features include Magic Pointer — point at anything on screen for Gemini to surface contextual suggestions — and Create My Widget, which generates custom widgets via AI.

SynthID Expands to More Platforms

One of the broader announcements with long-term implications is the expansion of SynthID, Google DeepMind's AI content watermarking system. Starting today, OpenAI, NVIDIA, and ElevenLabs are adopting SynthID to make their AI-generated video and audio content identifiable.

This matters for developers building applications that generate or display media content: SynthID provides a technical layer for identifying AI-generated material, which is becoming increasingly important as generative content becomes indistinguishable from human-created content.

Gemini Usage Numbers

Sundar Pichai shared a notable metric at the keynote: Gemini is currently processing 9.7 trillion tokens per month across all its products and API usage. That figure reflects both the scale of Google's AI deployment across Search, Workspace, and developer tools, and the growth of Gemini as a platform since its launch.

Where to Follow the Rest of the Event

The event continues through May 20. Google is streaming all technical sessions live at io.google. For developers, sessions on the Gemini API, Vertex AI, and Antigravity are the most relevant and will be available as recordings after the event.