Google I/O 2025: All AI Announcements in 32 Minutes

Google's AI Advancements

• 🚀 Ironwood, Google's 7th generation TPU, offers a 10x performance increase and 42.5 xlops of compute per pod, available to Google Cloud customers later this year.
• 💡 Google Beam, an AI-first video communication platform, uses six cameras and AI to render 3D light field displays for immersive conversations, with devices coming later this year in collaboration with HP.
• 🗣️ Real-time speech translation is now integrated into Google Meet, with English and Spanish available for subscribers and more languages rolling out soon.

Gemini and Agentic Capabilities

• 🤖 Gemini Live, an extension of Project Astra, allows users to share screens and discuss anything visible, rolling out on Android and iOS.
• 🧠 Project Mariner is an agent that can interact with the web, supporting multitasking with up to 10 simultaneous tasks and learning through a 'teach and repeat' feature, with its capabilities coming to developers via the Gemini API this summer.
• 🛠️ The Gemini SDK is now MCP compatible, enabling agentic capabilities in Chrome search and the Gemini app, including an experimental 'agent mode' for finding and scheduling tours for listings.
• 👤 Personal context allows Gemini models to use relevant Google app data privately and transparently with user permission, powering personalized smart replies that mimic the user's tone and style, coming to Gmail this summer.
• ⚡ Gemini Flash is an updated, efficient model with improved reasoning, code, and long context capabilities, available in preview now and generally available in early June.
• 🔊 New text-to-speech previews offer expressive, nuanced conversations with native audio output, supporting over 24 languages and seamless language switching, available today in the Gemini API.
• 🔒 Gemini 2.5 Pro and Flash include enhanced security against threats like indirect prompt injections and offer 'thought summaries' via the Gemini API and Vertex AI.
• 💰 Thinking Budgets are introduced for Gemini Flash to control cost and latency versus quality, with this feature rolling out to 2.5 Pro in the coming weeks.

AI in Search and Creative Tools

• 💻 Gemini 2.5 Pro can now update code based on image inputs and has native audio capabilities, with the asynchronous coding agent Jules now in public beta.
• 🎨 Gemini Diffusion generates images five times faster than previous models while matching coding performance.
• 🔬 Deep Think mode for Gemini 2.5 Pro is undergoing safety evaluations and will be available to trusted testers before wider release.
• 🌐 Gemini aims to become a 'world model' with capabilities like video understanding, screen sharing, and computer control, demonstrated through a demo of fixing a stripped screw and calling a bike shop.
• 🧬 Google's scientific AI research includes AlphaFold 3 for molecular structure prediction and Isomorphic Labs for AI-driven drug discovery.
• 🔍 AI Mode in Google Search offers an reimagined search experience with advanced reasoning for complex queries, rolling out in the US today, with personalized suggestions and connections to Gmail coming soon.
• 📊 Deep Search provides expert-level, cited reports by reasoning across disparate information, with complex analysis and data visualization coming this summer for sports and financial questions.
• 🎟️ Project Mariner's agentic capabilities are integrated into AI Mode for finding event tickets and completing checkouts.
• 🛍️ Search Live brings Project Astra's live capabilities into AI Mode for a video call-like search experience, including a virtual try-on feature for clothes and agentic checkout with Google Pay.
• 📄 Canvas transforms reports into dynamic web pages, infographics, quizzes, or podcasts, with options to 'vibe' content for various outputs.
• 🖼️ Imagen 4 is a new image generation model in the Gemini app, excelling at text and typography, and is 10 times faster than previous models.
• 🎬 VO3 is a new model with native audio generation for sound effects and dialogue, and Lyria 2 generates high-fidelity music and professional-grade audio.
• 💧 Synth ID detector can now identify AI-generated images, audio, text, or video with embedded watermarks.
• 🎥 Flow, a new AI filmmaking tool, allows creators to upload images, generate them with Imagine, maintain character and scene consistency, and extend or trim clips.

Google AI Subscriptions and XR

• 💰 Google AI Pro and Ultra subscription plans offer enhanced AI product suites, higher rate limits, and earlier access to new features.
• 👓 Android XR is being built for emerging form factors like headsets and glasses, optimized with Samsung and Qualcomm.
• 🕶️ Gemini on headsets is demonstrated with Samsung's Project Muhan, offering an infinite screen and integration with Google Maps.
• 👓 Android XR glasses are lightweight for all-day wear, with cameras, microphones, speakers, and optional in-lens displays for hands-free AI interaction, demonstrated backstage at I/O with live translation capabilities.
• 🏆 Gemini leads the AI counter leaderboard at 95.