Google I/O 2025 Unveils Cutting-Edge Tech Transforming Research Into Real-World Solutions

Note from the editor: This transcript captures highlights from Google CEO Sundar Pichai’s keynote at Google I/O 2025, expanded to cover additional announcements from the event. Explore all updates in our featured collection.
Traditionally, Google remained quiet in the weeks before I/O as we reserved major announcements for the stage. However, in the Gemini era, breakthroughs like AlphaEvolve or next-generation models could debut anytime—even on an ordinary Tuesday in March.
Our priority is delivering cutting-edge AI models into users' hands through our products faster than ever.
![]()
Unprecedented model advancements
The accelerated pace of AI model development is remarkable. Since Gemini Pro's first generation, our Elo scores—a key performance metric—have surged over 300 points. The current Gemini 2.5 Pro now dominates every category on the LMArena benchmark.
This progress is powered by our industry-leading infrastructure. Our seventh-gen TPU, Ironwood—the first architecture engineered specifically for large-scale AI reasoning—delivers 10x greater performance than its predecessor, achieving an extraordinary 42.5 exaflops per pod.
Our hardware advantage enables us to significantly reduce costs while maintaining performance leadership. Google doesn't just lead the Pareto Frontier—we're actively redefining it through continuous innovation.
![]()
Global AI adoption accelerates
AI capabilities are becoming universally accessible at an unprecedented rate. Key adoption metrics tell the story:
- Monthly token processing has skyrocketed from 9.7 trillion last year to over 480 trillion today—a 50x increase
- More than 7 million developers now build with Gemini (5x growth YoY), while Vertex AI usage has grown 40-fold
- The Gemini app reaches 400 million monthly active users, with 2.5 Pro adoption driving 45% higher engagement
Turning research into real-world impact
We're witnessing AI's transformational shift from theoretical research to practical applications benefiting individuals, businesses, and communities worldwide.
Project Starline evolves into Google Beam with speech translation
Our breakthrough 3D video technology, first showcased as Project Starline several I/O events ago, aimed to recreate the experience of shared physical presence.
Today we introduce Google Beam—an AI-powered video platform transforming standard 2D streams into immersive 3D experiences. Utilizing six cameras and advanced AI processing, Beam delivers millimeter-precise head tracking at 60fps for natural, real-time interactions. HP will debut the first commercial Beam systems later this year.
![]()
We're enhancing Google Meet with speech translation that preserves speakers' voices, tones, and expressions across languages. Initially launching for English/Spanish in beta for Pro/Ultra subscribers, this innovation will expand to Workspace enterprise users during 2025.
Project Astra becomes Gemini Live
Building on Project Astra's vision of contextual AI assistance, Gemini Live now incorporates camera and screen-sharing functionality—already helping users with diverse tasks from interview prep to fitness training. Following Android availability, iOS rollout begins today.
Project Mariner powers Agent Mode
AI agents combine model intelligence with tool access to complete tasks autonomously while remaining under user control.
Project Mariner—our prototype demonstrating web interaction capabilities—has advanced significantly since December. Its new "teach and repeat" system learns from single demonstrations to handle related future tasks. Partners like Automation Anywhere are already testing the Gemini API integration, with broader availability coming this summer.
This represents just one component of the emerging agent ecosystem infrastructure.
Initiatives like our open Agent2Agent Protocol (enabling inter-agent communication) and compatibility with Anthropic's Model Context Protocol exemplify this progress. Today we announce Gemini API/SDK support for MCP tools.
Early agent capabilities are coming to Chrome, Search, and Gemini. For example, Agent Mode helps simplify apartment hunting by analyzing Zillow listings, adjusting filters, and scheduling viewings—boosting conversion rates for partners while saving users hours.
Personalized AI experiences
Bridging research and everyday utility requires deep personalization. Our personal context system—private, transparent, and user-controlled—will connect relevant information across Google services.
Coming later this year, personalized Smart Replies in Gmail will harness past emails and Drive documents (like Docs itineraries) to craft responses reflecting your unique style. This exemplifies how personal context will transform Search, Gemini, and beyond.
Smarter search with AI Mode
Gemini models are making Search more intelligent, proactive, and tailored.
AI Overviews now serve 1.5 billion users globally, driving over 10% query growth in major markets. This represents Search's most successful launch in a decade.
For those wanting full AI-powered search, our redesigned AI Mode handles complex, multi-turn queries—with early testers submitting searches 2-3x longer than traditional queries. Rolling out today in the U.S., AI Mode features our fastest Gemini models yet.
Creative breakthroughs with Veo 3 and Imagen 4
Our newest generative media models—Veo 3 (now with audio generation) and Imagen 4 (our most advanced image model)—are coming to Gemini, empowering creators.
Building AI for everyone
This represents AI's greatest opportunity—to improve billions of lives through ongoing breakthroughs from robotics to quantum computing and autonomous vehicles.
A recent Waymo ride with my octogenarian parents reinforced how technology continues to inspire wonder while solving real problems—a reminder of why we do this work.
Related article
Warner Music acquires AI attribution startup Sureel AI
Warner Music Group (WMG) confirmed on Wednesday that it is acquiring Sureel AI, an artificial intelligence attribution startup. Sureel’s proprietary technology generates “AI DNA” for musical tracks, deconstructing them into constituent elements to tr
Amazon introduces Alexa for Shopping while pushing Rufus to the background
Amazon has launched Alexa for Shopping, merging its Rufus shopping chatbot with Alexa+ across the app, website, and Echo Show devices.The assistant answers product queries, compares items, tracks prices, and supports shopping reminders. It also handl
Microsoft, Azure and AI Tech Combat California Wildfire Risks
Microsoft invests in AI-driven wildfire detection, with Juan Lavista Ferres, CVP and Chief Data Scientist, discussing strategies to mitigate environmental damage.According to NASA, climate change impacts everyone on Earth, manifesting as rising tempe
Related Special Topic Recommendations
Comments (3)
0/500
Уже интересно, что покажет Google на этот раз. Особенно надеюсь, что их ИИ станет более доступным и менее 'лабораторным', а то иногда кажется, что разработки так и остаются на этапе прототипов... 🤔 Ну, а пока ждём — посмотрим на реальные кейсы.
I'm glad to see Google putting research into practical solutions. The 'cutting-edge' tech sounds exciting, but part of me wonders if this just ties us deeper into their ecosystem. It feels like the real innovation is in how they package and sell it. Still curious about the specifics!

Note from the editor: This transcript captures highlights from Google CEO Sundar Pichai’s keynote at Google I/O 2025, expanded to cover additional announcements from the event. Explore all updates in our featured collection.
Traditionally, Google remained quiet in the weeks before I/O as we reserved major announcements for the stage. However, in the Gemini era, breakthroughs like AlphaEvolve or next-generation models could debut anytime—even on an ordinary Tuesday in March.
Our priority is delivering cutting-edge AI models into users' hands through our products faster than ever.
Unprecedented model advancements
The accelerated pace of AI model development is remarkable. Since Gemini Pro's first generation, our Elo scores—a key performance metric—have surged over 300 points. The current Gemini 2.5 Pro now dominates every category on the LMArena benchmark.
This progress is powered by our industry-leading infrastructure. Our seventh-gen TPU, Ironwood—the first architecture engineered specifically for large-scale AI reasoning—delivers 10x greater performance than its predecessor, achieving an extraordinary 42.5 exaflops per pod.
Our hardware advantage enables us to significantly reduce costs while maintaining performance leadership. Google doesn't just lead the Pareto Frontier—we're actively redefining it through continuous innovation.
Global AI adoption accelerates
AI capabilities are becoming universally accessible at an unprecedented rate. Key adoption metrics tell the story:
- Monthly token processing has skyrocketed from 9.7 trillion last year to over 480 trillion today—a 50x increase
- More than 7 million developers now build with Gemini (5x growth YoY), while Vertex AI usage has grown 40-fold
- The Gemini app reaches 400 million monthly active users, with 2.5 Pro adoption driving 45% higher engagement
Turning research into real-world impact
We're witnessing AI's transformational shift from theoretical research to practical applications benefiting individuals, businesses, and communities worldwide.
Project Starline evolves into Google Beam with speech translation
Our breakthrough 3D video technology, first showcased as Project Starline several I/O events ago, aimed to recreate the experience of shared physical presence.
Today we introduce Google Beam—an AI-powered video platform transforming standard 2D streams into immersive 3D experiences. Utilizing six cameras and advanced AI processing, Beam delivers millimeter-precise head tracking at 60fps for natural, real-time interactions. HP will debut the first commercial Beam systems later this year.
We're enhancing Google Meet with speech translation that preserves speakers' voices, tones, and expressions across languages. Initially launching for English/Spanish in beta for Pro/Ultra subscribers, this innovation will expand to Workspace enterprise users during 2025.
Project Astra becomes Gemini Live
Building on Project Astra's vision of contextual AI assistance, Gemini Live now incorporates camera and screen-sharing functionality—already helping users with diverse tasks from interview prep to fitness training. Following Android availability, iOS rollout begins today.
Project Mariner powers Agent Mode
AI agents combine model intelligence with tool access to complete tasks autonomously while remaining under user control.
Project Mariner—our prototype demonstrating web interaction capabilities—has advanced significantly since December. Its new "teach and repeat" system learns from single demonstrations to handle related future tasks. Partners like Automation Anywhere are already testing the Gemini API integration, with broader availability coming this summer.
This represents just one component of the emerging agent ecosystem infrastructure.
Initiatives like our open Agent2Agent Protocol (enabling inter-agent communication) and compatibility with Anthropic's Model Context Protocol exemplify this progress. Today we announce Gemini API/SDK support for MCP tools.
Early agent capabilities are coming to Chrome, Search, and Gemini. For example, Agent Mode helps simplify apartment hunting by analyzing Zillow listings, adjusting filters, and scheduling viewings—boosting conversion rates for partners while saving users hours.
Personalized AI experiences
Bridging research and everyday utility requires deep personalization. Our personal context system—private, transparent, and user-controlled—will connect relevant information across Google services.
Coming later this year, personalized Smart Replies in Gmail will harness past emails and Drive documents (like Docs itineraries) to craft responses reflecting your unique style. This exemplifies how personal context will transform Search, Gemini, and beyond.
Smarter search with AI Mode
Gemini models are making Search more intelligent, proactive, and tailored.
AI Overviews now serve 1.5 billion users globally, driving over 10% query growth in major markets. This represents Search's most successful launch in a decade.
For those wanting full AI-powered search, our redesigned AI Mode handles complex, multi-turn queries—with early testers submitting searches 2-3x longer than traditional queries. Rolling out today in the U.S., AI Mode features our fastest Gemini models yet.
Creative breakthroughs with Veo 3 and Imagen 4
Our newest generative media models—Veo 3 (now with audio generation) and Imagen 4 (our most advanced image model)—are coming to Gemini, empowering creators.
Building AI for everyone
This represents AI's greatest opportunity—to improve billions of lives through ongoing breakthroughs from robotics to quantum computing and autonomous vehicles.
A recent Waymo ride with my octogenarian parents reinforced how technology continues to inspire wonder while solving real problems—a reminder of why we do this work.
Warner Music acquires AI attribution startup Sureel AI
Warner Music Group (WMG) confirmed on Wednesday that it is acquiring Sureel AI, an artificial intelligence attribution startup. Sureel’s proprietary technology generates “AI DNA” for musical tracks, deconstructing them into constituent elements to tr
Microsoft, Azure and AI Tech Combat California Wildfire Risks
Microsoft invests in AI-driven wildfire detection, with Juan Lavista Ferres, CVP and Chief Data Scientist, discussing strategies to mitigate environmental damage.According to NASA, climate change impacts everyone on Earth, manifesting as rising tempe
Уже интересно, что покажет Google на этот раз. Особенно надеюсь, что их ИИ станет более доступным и менее 'лабораторным', а то иногда кажется, что разработки так и остаются на этапе прототипов... 🤔 Ну, а пока ждём — посмотрим на реальные кейсы.
I'm glad to see Google putting research into practical solutions. The 'cutting-edge' tech sounds exciting, but part of me wonders if this just ties us deeper into their ecosystem. It feels like the real innovation is in how they package and sell it. Still curious about the specifics!





Home






