Google I/O ‘25 Keynote — full

Year: 2025 · ▶ Watch on YouTube

Sundar Pichai (CEO) · Demis Hassabis (CEO) · Tulsee Doshi (Head of Product, Gemini API) · Liz Reid (Head of Google Search) · Rajan Patel (VP, Engineering, Google Search) · Vidhya Srinivasan (VP/GM, Search Ads) · Josh Woodward (VP, Google Labs) · Jason Baldridge (Director, Generative Media) · Shahram Izadi (Director of Research, AR) · Nishtha (Product Manager)

Switch language → zh

Segments (10)

  • 01:15 · Opening & The Gemini Era Progress — Sundar Pichai
    • Sundar Pichai kicks off the keynote, highlighting the rapid pace of AI development and shipping in the ‘Gemini era’, showcasing significant progress in model performance and adoption.
  • 18:24 · Google DeepMind & The Future of AI — Demis Hassabis
    • Demis Hassabis discusses the progress towards AGI, introducing new model capabilities like Deep Think, world models, and advancements in AI for science with AlphaFold 3 and AlphaEvolve.
  • 22:16 · Gemini 2.5 for Developers — Tulsee Doshi
    • Tulsee Doshi details improvements to the Gemini 2.5 models for developers, including enhanced security, better cost-efficiency, native audio output, and the introduction of the coding agent, Jules.
  • 47:50 · The New Era of Google Search — Liz Reid
    • Liz Reid unveils the future of Google Search, introducing AI Mode, Deep Search, and personalized suggestions powered by Gemini 2.5 to handle more complex and personal queries.
  • 54:47 · Complex Analysis & Agentic Search — Rajan Patel
    • Rajan Patel demonstrates advanced data analysis, visualization, and agentic capabilities within AI Mode, showing how Search can now perform multi-step tasks like finding tickets.
  • 01:01:00 · Shopping in the AI Era — Vidhya Srinivasan
    • Vidhya Srinivasan showcases how AI is transforming shopping in Search with visual inspiration, virtual try-on features, and agentic checkout capabilities.
  • 01:09:57 · The Universal AI Assistant: Gemini App — Josh Woodward
    • Josh Woodward outlines the vision for Gemini as a personal, proactive, and powerful universal AI assistant, introducing Gemini Live, Deep Research, Canvas, and Gemini in Chrome.
  • 01:22:42 · Generative Media & Creative Tools — Jason Baldridge
    • Jason Baldridge introduces new generative media tools, including Music AI Sandbox with Lyria 2, and partnerships with filmmakers to develop Veo as a professional storytelling tool.
  • 01:36:30 · Android & The Physical World — Shahram Izadi
    • Shahram Izadi presents Android XR, a new platform for headsets and glasses built in the Gemini era, and demonstrates Project Astra’s real-time, multimodal capabilities on prototype glasses.
  • 01:50:08 · Closing Remarks & Vision for AI — Sundar Pichai
    • Sundar Pichai concludes the keynote by summarizing the announcements and reinforcing Google’s mission to make AI helpful for everyone, highlighting real-world applications like wildfire detection and disaster relief.

Products Announced (16)

  • 01:55 · Gemini 2.5 Pro (Updated Version)
    • State-of-the-art performance across all LMArena categories · Improved coding capabilities, #1 on WebDev Arena · Powers AI Mode in Google Search
    • Available in Gemini App and Search
  • 07:32 · Google Beam (New Product)
    • AI-first 3D video communication platform · Uses a new video model to transform 2D streams into a realistic 3D experience · Developed in partnership with HP
    • Coming to early customers later this year
  • 10:08 · Google Meet Speech Translation (New Feature)
    • Real-time speech translation directly in Google Meet · Matches speaker’s tone, patterns, and expressions · Initially supports English and Spanish
    • Available now for subscribers, more languages in coming weeks
  • 10:46 · Gemini Live (New Feature in Gemini App)
    • Real-time conversational experience · Includes camera and screen sharing capabilities from Project Astra · Allows users to talk about anything they can see
    • Rolling out on Android and iOS starting today
  • 12:24 · Agent Mode in Gemini App (Experimental)
    • Performs multi-step tasks on the user’s behalf · Can interact with the web and other services (e.g., Zillow) · Uses Project Mariner capabilities
    • Coming soon to subscribers
  • 16:15 · Personalized Smart Replies in Gmail (New Feature)
    • Generates email replies that sound like the user · Uses personal context from Drive, past emails, and Docs · Matches user’s tone, style, and word choices
    • Available in Gmail this summer for subscribers
  • 20:02 · Gemini 2.5 Flash (Updated Version)
    • More efficient and cost-effective model · Improved across reasoning, code, and long context · Ranks second only to 2.5 Pro on LMArena
    • Generally available in early June
  • 22:42 · Gemini Text-to-Speech (New Previews)
    • Native audio output for more expressive voices · Multi-speaker support for two voices · Works in over 24 languages
    • Available in the Gemini API today
  • 30:08 · Jules (Public Beta)
    • Asynchronous coding agent · Fixes bugs, makes updates, and integrates with GitHub · Can handle complex tasks in large codebases
    • Public beta available at jules.google
  • 31:37 · Gemini Diffusion (Experimental Research Model)
    • Text diffusion model for extremely low-latency generation · Generates 5x faster than 2.0 Flash-Lite · Excels at editing tasks for math and code
    • Currently in testing with a small group
  • 46:50 · AI Mode in Google Search (New Feature)
    • End-to-end AI search experience · Handles longer, more complex conversational queries · Features multi-step reasoning and planning
    • Rolling out to everyone in the U.S. starting today
  • 01:03:23 · Virtual Try-On in Google Search (New Feature)
    • Virtually try on clothes using a personal photo · Powered by a custom image generation model for fashion · Shows how material will drape, fold, and stretch
    • Rolling out in Labs beginning today
  • 01:17:33 · Imagen 4 (New Model)
    • Most capable image generation model · Improved text and typography rendering · 10x faster variant available
    • Available in the Gemini app starting today
  • 01:19:34 · Veo 3 (New Model)
    • State-of-the-art video generation model · Includes native audio generation (sound effects, dialogue) · Improved photorealism and physics understanding
    • Available today
  • 01:30:27 · Flow (New AI Filmmaking Tool)
    • Combines Veo, Imagen, and Gemini · Built for creatives, allows for scene building and iteration · Maintains character and scene consistency across clips
    • Launching today
  • 01:36:34 · Android XR (New Platform)
    • Platform for immersive headsets and glasses · Built in the Gemini era for AI-first experiences · Developed in partnership with Samsung and Qualcomm
    • Developer preview available, first devices later this year

Benchmarks Shown (7)

  • 02:25 · Model Progress (Debut LMArena Elo Score): 1448
    • Shows step-function increase from Gemini 1.0 Pro (1111) to Gemini 2.5 Pro (1448).
  • 02:38 · LMArena Leaderboard: 1
    • Gemini 2.5 Pro is #1 across all categories (Overall, Hard Prompts, Coding, Math, etc.).
  • 02:55 · WebDev Arena: +142
    • Updated Gemini 2.5 Pro shows a +142 Elo score increase vs. the March release.
  • 04:35 · LMArena Leaderboard (Fastest Intelligence): 332
    • Google models (Gemini 2.5 Flash, x3, Gemini 2.0 Flash) hold the top 3 spots for speed.
  • 33:33 · USAMO 2025: 49.4%
    • Gemini 2.5 Pro Deep Think significantly outperforms Gemini 2.5 Pro (34.5%) and other models on this math benchmark.
  • 33:33 · LiveCodeBench v6: 80.4%
    • Gemini 2.5 Pro Deep Think leads over other models in this coding benchmark.
  • 33:33 · MMMU: 84.0%
    • Gemini 2.5 Pro Deep Think shows top performance in multimodality benchmark.

Commitments / Timelines (11)

  • 07:32 (later this year) — Google Beam devices will be available for early customers.
  • 10:08 (in the coming weeks) — More languages for Google Meet speech translation will be rolling out.
  • 10:46 (starting today) — Gemini Live with camera and screen sharing is rolling out.
  • 12:24 (coming soon) — An experimental version of Agent Mode will come to the Gemini app for subscribers.
  • 16:15 (this summer) — Personalized Smart Replies will be available in Gmail for subscribers.
  • 20:02 (in early June) — Gemini 2.5 Flash will be generally available.
  • 30:43 (today) — Jules, the asynchronous coding agent, is now in public beta.
  • 46:50 (starting today) — AI Mode in Google Search is rolling out to everyone in the U.S.
  • 01:03:23 (beginning today) — Virtual Try-On feature is available in Labs.
  • 01:16:40 (this week) — Gemini in Chrome is rolling out for Gemini subscribers in the US.
  • 01:49:02 (later this year) — Developers can start developing for Android XR glasses.

Demos (6)

  • 03:30 ✓ · Pokémon Blue Playthrough — Sundar Pichai (narrating)
    • A progress timeline chart showing Gemini completing all major milestones in the game Pokémon Blue over 700 hours.
  • 08:57 ✓ · Google Meet Real-time Speech Translation — Pre-recorded actors
    • Two people speaking different languages (English and Spanish) have a fluid conversation in Google Meet with real-time, voice-matched translation.
  • 11:08 ✓ · Gemini Live with Camera (Object Identification) — Pre-recorded user
    • A user points their phone camera at various objects (garbage truck, street light, their own shadow) and Gemini Live correctly identifies them, humorously correcting the user’s misidentifications.
  • 01:03:34 ✓ · Virtual Try-On Live Demo — Vidhya Srinivasan
    • Vidhya uses her phone to take a photo of herself and then virtually tries on a dress, showing the generated image of her wearing the dress on the screen.
  • 01:41:46 ✓ · Android XR Glasses Live Demo — Shahram Izadi and Nishtha
    • A live, on-stage demo where Nishtha, wearing prototype glasses, uses Gemini to identify people and objects backstage, get information, and have a real-time translated conversation with Shahram.
  • 25:55 ✓ · AI Studio 3D Web App Generation — Tulsee Doshi
    • Tulsee demos how Gemini 2.5 Pro in AI Studio can take a hand-drawn sketch of a 3D photo sphere and generate the corresponding interactive web application code (HTML, CSS, JS).

Notable Quotes (8)

  • 01:37 — Sundar Pichai:

    Every day is Gemini season here at Google.

  • 02:07 — Sundar Pichai:

    And so we are shipping faster than ever.

  • 03:48 — Sundar Pichai:

    Artificial Pokémon Intelligence.

  • 06:38 — Sundar Pichai:

    Google Search is bringing generative AI to more people than any other product in the world.

  • 18:30 — Demis Hassabis:

    We’re living through a remarkable moment in history, where AI is making possible an amazing new future.

  • 48:40 — Liz Reid:

    Today you’ll see how you can ask anything.

  • 01:10:03 — Josh Woodward:

    One that doesn’t just respond, but understands. One that doesn’t just wait, but anticipates.

  • 01:53:44 — Sundar Pichai:

    It was a reminder of how incredible the power of technology is to inspire, to awe, and to move us forward.

Visual Signals (Beyond the Transcript)

On-Screen Text Moments (7)

  • 00:06 · A satellite shaped like the number 10 orbiting Earth.
    • This is the opening visual of the keynote’s countdown, creatively generated by AI to represent the number 10.
  • 02:08 · A timeline titled 'Shipping at Relentless Pace' showing dozens of model and product releases since t
    • Visually reinforces Google’s key message about their accelerated pace of innovation and shipping in the AI space.
  • 02:25 · A bar chart titled 'Model Progress' showing the steep increase in LMArena Elo scores from Gemini 1.0
    • Provides a clear, quantitative visualization of the rapid improvement in their AI model capabilities over 18 months.
  • 04:51 · A Pareto Frontier graph showing Google's models occupying the optimal top-left quadrant for performa
    • Graphically argues that Google provides the best performance at the most effective price point, and is pushing the entire frontier of what’s possible.
  • 05:20 · A line graph showing 'Monthly Tokens Processed' skyrocketing from 9.7T to 480T+ in one year.
    • Dramatically illustrates the massive 50x increase in AI adoption and usage across Google’s products and APIs.
  • 33:10 · The title card 'Gemini 2.5 Pro Deep Think'.
    • Introduces a new, more powerful reasoning mode for their top model, signaling a focus on deeper, more complex problem-solving.
  • 01:13:05 · The words 'Personal, Proactive, Powerful' on screen.
    • These three words define Google’s core vision for its universal AI assistant, Gemini.

Stage Moments (7)

  • 01:14 · Sundar Pichai walks onto the large, circular stage in front of a massive audience in an outdoor amphitheater.
  • 02:43 · Audience applauds loudly for the Gemini 2.5 Pro benchmark results.
  • 18:11 · Sundar Pichai welcomes Demis Hassabis to the stage with a hug.
  • 47:32 · The audience gives a strong, sustained applause for the announcement that AI Mode is rolling out in the US.
  • 01:43:50 · Nishtha walks on stage wearing the prototype Android XR glasses and interacts with Shahram Izadi.
  • 01:46:04 · Giannis Antetokounmpo makes a surprise cameo appearance in the Android XR glasses demo, high-fiving Nishtha.
  • 01:53:53 · Sundar Pichai returns to the stage for closing remarks, followed by a final sizzle reel of community creations.

Visual Demos (5)

  • 00:06 · AI-generated video countdown
    • A montage of visually stunning, surreal, and photorealistic video clips generated by AI, each creatively incorporating a number from 10 down to 1.
  • 01:30:35 · Flow AI Filmmaking Tool Demo
    • A user interface for ‘Flow’ is shown, where a user combines images and text prompts to generate a sequence of video clips, including a flying car with a giant chicken on it.
  • 01:28:07 · Ancestra Short Film Trailer
    • A trailer for a short film by Eliza McNitt, executive produced by Darren Aronofsky, showcasing a mix of live-action and surreal, cosmic, and microscopic visuals generated by Veo.
  • 37:42 · Project Astra Prototype Glasses Demo
    • A pre-recorded, first-person view from a user wearing prototype glasses, showing the AI assistant identifying objects, remembering locations, and controlling on-screen elements in real-time.
  • 59:44 · Search Live with Video Demo
    • A montage of users pointing their phone cameras at real-world objects (science experiments, plants, remote controls) and having a live, conversational search experience with Gemini.

Production Signals (7)

  • 00:00 · Pre-recorded AI-generated video intro
  • 01:14 · Transition to live on-stage presentation
  • 08:57 · Pre-recorded demo segment (Google Meet Translation)
  • 37:42 · Pre-recorded demo segment (Project Astra)
  • 59:44 · Pre-recorded demo segment (Search Live)
  • 01:03:34 · Live on-stage demo (Virtual Try-On)
  • 01:41:46 · Live on-stage demo (Android XR Glasses)

Key Topics

Generative AI · Gemini Model Family · Multimodality · AI Agents · Google Search · AI Overviews · Developer Tools · Creative AI Tools · Video Generation (Veo) · Image Generation (Imagen) · Android XR · AI Assistants · Personalization · AI for Science

Takeaways

  • Google is all-in on the ‘Gemini Era,’ integrating its most advanced AI models across its entire product ecosystem, from Search and Android to creative and developer tools.
  • The future of Google’s products is agentic, personal, and proactive; AI will not just respond to queries but anticipate needs and perform multi-step tasks on the user’s behalf.
  • Multimodality is central to Google’s strategy, with a heavy focus on real-time conversational AI that can see, hear, and speak, demonstrated through Gemini Live and the Project Astra glasses prototype.
  • Google is rapidly shipping new models (Gemini 2.5 Pro/Flash, Veo 3, Imagen 4) and features, emphasizing speed and making state-of-the-art capabilities available to developers and users ‘today’ or ‘soon’.
  • Search is being fundamentally reimagined with ‘AI Mode,’ moving beyond simple answers to become a comprehensive research and planning partner capable of deep analysis and visualization.
  • A new hardware frontier is opening with Android XR, a platform for glasses and headsets that will serve as a natural interface for a persistent, context-aware AI assistant.
  • Google is heavily investing in generative media, providing powerful tools like Veo, Imagen, and Lyria to empower creators and filmmakers, blurring the line between prompting and professional production.