A message from Sundar Pichai, CEO of Google and Alphabet: Central to human progress is information, and for over 26 years, we've been dedicated to organizing global knowledge to make it both accessible and useful. Our ongoing effort to harness AI aims to manage information through various inputs and outputs, ensuring it’s genuinely beneficial for you. This vision led to the launch of Gemini 1. 0 last December, our first naturally multimodal model. Gemini 1. 0 and 1. 5 marked significant leaps in multimodality and long context, allowing us to comprehend and process information across text, video, images, audio, and code more effectively. Currently, millions of developers use Gemini to innovate our products, including those with 2 billion users, and to develop new ones. An example is NotebookLM, showcasing the benefits of multimodality and long context, making it a favorite for many. In the past year, we've focused on creating more agentic models capable of understanding the world around you, anticipating future steps, and acting with your guidance. Today, we’re thrilled to unveil Gemini 2. 0, our most advanced model yet, designed for this new agentic era. With breakthroughs in multimodality, like native image and audio output, and built-in tool use, it paves the way for developing new AI agents aligned with our vision of a universal assistant. We are deploying 2. 0 to developers and trusted testers now and swiftly integrating it into our products, beginning with Gemini and Search. From today, the Gemini 2. 0 Flash experimental model will be open to all Gemini users.

Additionally, we’re unveiling Deep Research, a feature that acts as a research assistant by delving into complex topics and creating reports, available in Gemini Advanced today. No product has been more AI-transformed than Search. Our AI Overviews reach 1 billion people, allowing new types of inquiries and rapidly becoming a favored feature. Next, the advanced reasoning power of Gemini 2. 0 will enhance AI Overviews to handle intricate topics and multi-step questions, including sophisticated math, multimodal queries, and coding. Limited testing began this week, with a broader rollout planned for early next year, extending AI Overviews to more countries and languages. The progress with 2. 0 is underpinned by long-term investments in our distinctive full-stack AI innovation approach. Built on custom hardware like Trillium, our sixth-generation TPUs, these TPUs powered all Gemini 2. 0 training and inference. Now, Trillium is generally available to customers for their own builds.