100 things we announced at I/O 2026

Source: Google Date: 2026-05-20 URL: https://blog.google/innovation-and-ai/technology/ai/google-io-2026-all-our-announcements/

Summary

Google I/O 2026 headline items: Gemini 3.5 Flash (new frontier model, claims better-than-3.1-Pro coding at sub-half-cost), Gemini Omni (multimodal-in to any output, starting with video), and Google Antigravity 2.0 (multi-agent orchestration desktop app with CLI + SDK). The API now provisions managed remote Linux sandboxes where agents browse, execute code, and manage files in isolation. Chrome DevTools gained agent-debugging visibility. WebMCP was proposed as an open standard for exposing browser-native tools to agents. On the consumer side: AI Mode in Search (conversational, Gemini 3.5 Flash), Search Agents (autonomous 24/7 web monitoring), and Gemini for Science (hypothesis generation, literature synthesis across 30+ life science databases).

Implications

• Agentic engineering: Antigravity 2.0 + Managed Agents API is Google’s clearest move onto the orchestration layer — remote sandboxed environments provisioned per API call is infrastructure-as-agent-substrate. WebMCP as an open standard is an attempt to establish a browser-native tool protocol; if it gains traction it competes with MCP (Anthropic) at the browser boundary. Chrome DevTools integration for agent debugging is a tooling primitive that doesn’t exist yet in any competing stack.
• Coding agents: Gemini 3.5 Flash’s coding benchmark claims and the AI Studio → Google Play Console pipeline position Google as a credible coding-agent provider, not just a model vendor. This is a direct threat to Cursor/Windsurf/Codex CLI’s model-agnostic positioning.
• AI ecosystem/power dynamics: The breadth of I/O 2026 — models, orchestration infra, consumer AI, science tools, ad creative, design agents — is Google executing on every layer simultaneously. The cost angle on Gemini 3.5 Flash (“less than half the cost of other frontier models”) is a pricing-pressure signal aimed directly at Anthropic and OpenAI margins.
• Local models: Notably absent from the I/O summary: on-device or local model announcements. Google’s I/O bet is cloud-hosted agent infrastructure, not edge deployment — a strategic contrast with Apple’s on-device push and the open-weight ecosystem.