Google has used its annual I/O 2025 developer conference to unveil a sweeping overhaul of its artificial intelligence (AI) portfolio, marking a huge leap in performance, product integration and accessibility. Chief executive Sundar Pichai outlined the company's “Gemini era” vision — characterised by relentless model updates, exponential infrastructure gains, and deeper AI personalisation across Google’s ecosystem.
AI shipping at speed
Gone are the days when Google saved its biggest AI announcements for the I/O stage. “We’re just as likely to ship our most intelligent model on a Tuesday in March,” said Pichai, highlighting the rapid pace of updates, including the release of Gemini 2.5 Pro, which now dominates the LMArena benchmark leaderboard.
This progress is underpinned by Ironwood, Google's seventh-generation Tensor Processing Unit (TPU), delivering a tenfold performance increase and enabling 42.5 exaflops of compute per pod. With this, Google says it has pushed forward the Pareto Frontier — improving both speed and cost-efficiency across AI workloads.
AI adoption accelerates
The global uptake of Google’s AI tools has been swift. In the past year, token processing volume has surged from 9.7 trillion to over 480 trillion tokens per month. More than 7 million developers are now building with Gemini, while the Gemini app has attracted over 400 million monthly active users. Use of Gemini 2.5 Pro within the app has increased 45% quarter-on-quarter.
From research to real-world use
Former experimental projects are now commercial products. Project Starline, first introduced as a breakthrough in 3D video calling, has evolved into Google Beam. The new platform uses AI and six-camera arrays to convert 2D video into real-time 3D conversations at 60 frames per second. Developed with HP, Google Beam will roll out to early customers later this year.
In parallel, speech translation in Google Meet will allow near real-time interpretation with matching voice and facial expression. English and Spanish are available to Google AI Pro and Ultra subscribers in beta, with further languages to follow.
Gemini Live and the universal assistant
Google’s vision for a context-aware AI assistant is also coming to life. Gemini Live, building on Project Astra, now incorporates camera and screen-sharing capabilities. Already live on Android and rolling out to iOS, the assistant supports a range of practical applications — from job interview prep to fitness coaching.
Agents and autonomous AI
Another milestone is the debut of Agent Mode, stemming from Project Mariner. These AI-powered agents can learn, perform tasks across the web, and adapt through “teach and repeat” interactions. Project Mariner’s capabilities are being extended to developers via the Gemini API, with initial partnerships including Automation Anywhere and UiPath.
Agent Mode is also being integrated into the Gemini app. For instance, while apartment hunting, the agent can search websites like Zillow, refine filters, and schedule inspections. An experimental version will be available to subscribers soon.
Google also announced compatibility between its Gemini API and Anthropic’s Model Context Protocol (MCP), enabling greater collaboration between AI agents.
Personalisation through context
Personalisation stands as a core theme in Google's AI push. With user consent, Gemini models can now use contextual information from Gmail and Google Drive to generate responses. A new feature — personalised Smart Replies in Gmail — will tailor messages based on past emails, user tone and preferences. This capability will become available to subscribers later in the year.
AI Mode transforms Search
Google Search is undergoing its most substantial evolution in a decade with the introduction of AI Mode, enabling end-to-end AI-powered search experiences. AI Overviews, already live for 1.5 billion users globally, are driving increased engagement and query volumes in key markets like the US and India.
AI Mode supports longer, more complex queries and follow-ups — reflecting a deeper level of reasoning. This feature is now rolling out to US users, alongside the introduction of Gemini 2.5 into Search.
Gemini 2.5: Smarter, faster, more efficient
The Gemini 2.5 family continues to evolve. Gemini 2.5 Flash, Google’s fast, low-cost variant, has seen strong uptake among developers. The flagship 2.5 Pro model now features a new Deep Think mode, incorporating cutting-edge parallel reasoning techniques for improved problem-solving.
Expanding the Gemini app ecosystem
Gemini is becoming a central productivity hub. Deep Research now supports uploads and soon Google Drive and Gmail integration. The AI assistant can produce detailed reports and generate infographics, quizzes and podcasts via the Canvas interface. Coding capabilities are also expanding through natural language interactions.
Gemini Live is now freely available across platforms, including iOS, and will soon connect seamlessly with other Google apps.
Generative media breakthroughs
Google also debuted Veo 3, its latest video generation model featuring native audio creation, and Imagen 4, an advanced image generation tool. These tools, embedded in the Gemini app, are poised to drive new levels of creativity. The company also introduced Flow, a filmmaker-focused tool for cinematic video generation and clip extension.
A new AI-powered era
Pichai’s address closed with a clear message: the AI platform shift is no longer speculative — it is here. With an accelerated release cadence, multi-modal intelligence, and deeper integration into daily tools and workflows, Google is reshaping its ecosystem around AI.
As Pichai noted, “We’re in a new phase where decades of research are now becoming reality.”