Google’s biggest developer conference is Google I/O 2025, which takes place Tuesday and Wednesday at the Shoreline Amphitheatre in Mountain View. It showcases AI advancements and a collaborative future in tech. At Google I/O 2025, CEO Sundar Pichai introduced groundbreaking AI technologies, including advanced AI assistants, improvements in Android, Chrome, Google Search, and innovative filmmaking tools.
Key takeaways of the Google I/O 2025 event
The following are all the things announced at Google I/O 2025.
Google’s 7th Generation TPU – Ironwood
Google has announced its 7th generation Tensor Processing Unit (TPU) “Ironwood,” which delivers 10 times more performance than the previous generation. It offers a compute capacity of 42.5 exalops per pod. It will be available to Google Cloud customers by the end of the year.
Google Beam: AI-First Video Communication Platform
Google has introduced a new AI-based video communication platform called “Google Beam.” In this, an array of six cameras captures the user from different angles and, with the help of AI, combines these video streams and renders them on a 3D light field display. It has head tracking with millimeter accuracy, and all this happens in real time at 60 frames per second. In collaboration with HP, Google Beam devices will be available to select customers by the end of this year.
Real-time speech translation in Google Meet
Starline’s technology is being added to Google Meet, including real-time speech translation. Users can now instantly see translations of conversations in English and Spanish, and support for more languages will be added in the coming weeks. The feature will also come to enterprise customers later this year.
Gemini Live and Project Astra capabilities
Gemini Live now offers camera and screen sharing so users can talk about any scene. For example, it can figure out if a thin building is actually a street light or if someone “following you” is just your shadow. The feature is available today on Android and iOS.
Project Mariner: Web interaction agent
Project Mariner is an agent that can multitask on the web. It can handle up to 10 tasks at a time and learn with a “teach and repeat” feature. These computer usage capabilities will be available to developers via the Gemini API this summer.
Gemini SDK is now compatible with MCP tools.
The Gemini SDK is now compatible with the Model Context Protocol (MCP) developed by Anthropic. Agentic capabilities are now being brought to the Gemini app and Chrome Search.
Agent Mode in the Gemini App
The new Agent Mode in the Gemini App lets users search for listings from sites like Zillow, filter using Project Mariner, and even schedule appointments through MCP. This Agent Mode will come to consumers soon.
Personal Context and Personalized Smart Replies
Gemini will now be able to respond to you in your own way, with your permission, using information from your Google Apps. This feature will be available in Gmail this summer.
Gemini Flash 2.5 and Text-to-Speech upgrades
The new Gemini Flash 2.5 is better at reasoning, code, and long context. It will be available in early June. Also, the new text-to-speech model now supports two voices and works in over 24 languages.
Gemini 2.5: Security and Thought Summaries
Gemini 2.5 is now the most secure model, including Thought Summaries. Thinking budgets also allows users to balance cost and quality.
2.5 Pro and Coding Agent Jules
Gemini 2.5 Pro is available in Android Studio, Firebase Studio, and Gemini Code Assist. Also, Coding Agent Jules is now in public beta.
Gemini Diffusion and Deep Think Mode
The new Gemini Diffusion model is 5 times faster at generating code from text. The Deep Think mode scores high in difficult math benchmarks like USA MO 2025.
Gemini: Towards a World Model
Gemini is now being worked on to become a universal AI assistant. This includes Project Astra’s capabilities, such as video understanding, screen sharing, and memory.
AI Uses in Life Sciences
Google has developed research systems such as Amy, AlphaFold 3, and Alpha Evolve that could revolutionize medicine and drug discovery.
AI Mode and Deep Search
Google has launched a new AI mode that better understands complex and long queries. It is available to everyone in the US starting today and will add Personal Context this summer.
Deep Search and Data Visualization in AI Mode
Deep Search gives users expert-level reports in just a few minutes, including graphs and tables.
Project Mariner and Astra’s capabilities in AI Mode
AI Mode is now adding agentic capabilities and the Search Live feature, allowing you to search like a live video call.
Visual Try-on and Agentic Checkout
With the new feature of Visual Shopping, users can try clothes virtually and make a purchase with one tap with the help of a checkout agent. This feature will be available soon.
Expansion of the Gemini Live and Canvas feature
Gemini Live can now connect to apps like Maps and Calendar. In Deep Research, users can now upload their files and convert them into websites, infographics, quizzes, etc., in one click
