Here is a breakdown of everything Google announced during the 2025 I/O keynote, captured in a quick, scannable format:

Core AI Infrastructure & Next-Gen Models

• Ironwood TPU (7th Gen): Google’s first hardware designed specifically for thinking and inferential AI workloads. It delivers 10x the performance of the previous generation and packs 42.5 exaflops per pod.

• Gemini 2.5 Pro & Deep Think: Features a native audio capability to explain its own thought processes out loud. The new Deep Think mode scored historic highs on the USA Mo 2025 math benchmark.

• Gemini 2.5 Flash: Built to be the high-efficiency workhorse model for mainstream application development.

• Gemini Diffusion: A highly optimized experimental text-diffusion model that generates images 5x faster than previous iterations while matching top-tier coding performance.

Developer Agents & Ecosystem

• Jules: An asynchronous, enterprise-grade AI coding assistant designed to tackle complex, multi-file tasks across massive codebases in minutes rather than hours.

• Model Context Protocol (MCP) Compatibility: The Gemini SDK is now fully compatible with MCP tools, allowing AI agents to plug seamlessly into external data sources and services.

• Project Mariner: A research prototype capable of autonomous web browsing, featuring multitasking and a “teach-and-repeat” setup for complex workflows.

Search & Consumer Apps

• AI Mode in Google Search: Introduces advanced reasoning, deep search pipelines for complex multi-step planning, and highly personalized results.

   

• Agentic Checkout: Enhances visual shopping. Users can utilize a virtual try-on feature for clothing and hand off the entire checkout and purchasing process to an AI agent.

• Canvas: A new Gemini workspace feature that transforms dense documents or reports into interactive web pages, infographics, flashcard quizzes, or custom audio podcasts in 45 languages with a single tap.

   

Creative & Media Tools

• Flow: A professional-grade AI filmmaking platform. Creators can anchor characters, styles, and scenes, then direct specific camera paths and movements using natural language instructions.

• Media Model Upgrades: Previews of Imagen 4 (10x faster layout and typography rendering), Veo 3 for high-fidelity video generation, and Lyria 2 for automated music composition and multi-language vocal arrangement.

Hardware, XR & Communication

• Google Beam: The evolution of Project Starline. It uses an array of six cameras and a new 3D video model to render a realistic lightfield experience on a standard display with millimeter-accurate head tracking at 60 FPS.

• Android XR & Glasses: An ecosystem built for lightweight, all-day smart glasses and headsets (including upcoming Samsung hardware). It offers an “infinite screen” workspace, real-time spatial translation, and deep integration with Gemini Live for camera and screen sharing.