MOUNTAIN VIEW, CA (Reporting from Toronto, ON) – May 26, 2025) – Google’s annual developer conference, I/O 2025, concluded today, firmly establishing artificial intelligence as the central pillar of the company’s future. A torrent of announcements revealed next-generation multimodal AI models, a fundamental reimagining of Google Search with “AI Mode,” and the pervasive embedding of its Gemini AI across its vast ecosystem of products and services.
“What all this progress means is that we’re in a new phase of the AI platform shift,” stated CEO Sundar Pichai, emphasizing that “decades of research are now becoming reality.”
New AI Powerhouses Unveiled
Google introduced several advanced AI models. Imagen 4, an image generation model, is now available on Vertex AI, promising superior image quality, text rendering, and speed. It succeeds Imagen 3, previously used for photorealistic image creation.
For video, Veo 3 was launched as a state-of-the-art generation model, uniquely featuring native support for sound effects, background noise, and dialogue. Also available on Vertex AI, Veo 3 is complemented by Flow, an AI-powered filmmaking tool.
Music generation sees the introduction of Lyria 2 on Vertex AI, alongside the experimental Lyria RealTime, an interactive music creation model accessible via the Gemini API. For on-device applications, Google presented Gemma 3n, an open multimodal model designed for efficiency on phones and laptops, handling audio, text, image, and video.
In a move towards responsible AI, Google launched the SynthID Detector, a verification portal to identify AI-generated content created with Google tools. This complements SynthID, which embeds imperceptible watermarks. For developers, Stitch, an experimental tool, aims to streamline the design-to-code process using AI.
Search Reimagined with AI Mode
Google Search is set for a significant transformation with the rollout of AI Mode in the U.S. Powered by a custom version of Gemini 2.5, it offers advanced reasoning, multimodality, and deeper follow-up capabilities. It utilizes a “query fan-out” technique for comprehensive answers.
AI Mode will soon feature Deep Search for more thorough, cited reports and personal context for suggestions based on user history (with user control emphasized). Complex data analysis and visualization for sports and finance are slated for this summer, alongside agentic capabilities from Project Mariner for task completion like booking reservations.
Gemini Model Family Expands and Evolves
The Gemini family of models saw significant updates. An improved Gemini 2.5 Flash is now the default in the Gemini app and will be generally available in June, followed by an updated Gemini 2.5 Pro, which now leads in coding benchmarks and learning capabilities. A new “Deep Think” mode for 2.5 Pro will tackle complex math and coding.
New capabilities for Gemini 2.5 Pro and Flash include native audio output, enhanced security, Project Mariner computer use features, and “thought summaries” for transparency. “Thinking budgets” are being extended to 2.5 Pro.
Beyond updates, new models include the aforementioned Gemma 3n, Gemini Diffusion for rapid text generation, and Lyria RealTime.
Gemini Integrated Across Google’s Landscape
The Gemini app itself gains Gemini Live (free with camera/screen sharing), built-in Imagen 4 and Veo 3 (Veo 3 for Ultra subscribers in the U.S.), and enhanced Deep Research and Canvas features. Interactive quizzes are also being added.
Gemini is also coming to Chrome on desktop for contextual assistance (rolling out to Pro/Ultra subscribers in the U.S.). Gemini Code Assist is now generally available, and Jules, an AI coding agent, is in public beta.
Workspace apps will see Gemini-powered personalized smart replies and inbox cleanup in Gmail, fast appointment scheduling, speech translation in Meet, and AI enhancements in Vids like transcript trimming and AI avatars.
New AI Subscription Tiers
Google clarified its AI subscription plans: Google AI Pro (formerly Gemini Advanced) offers premium features globally, with a free year for eligible college students. The new Google AI Ultra tier, launching in the U.S. and rolling out globally, provides the highest rate limits and early access to features like Deep Think, Flow with Veo 3, and 30TB of storage.
Developer tools like the Google GenAI SDK and an experimental URL context feature for the Gemini API were also announced. While some AI shopping features were met with internal skepticism by observers, the overarching message was one of profound AI integration.