A new AI model for the agentic era

A observe from Google and Alphabet CEO Sundar Pichai:

Data is on the core of human progress. It’s why we’ve targeted for greater than 26 years on our mission to prepare the world’s data and make it accessible and helpful. And it’s why we proceed to push the frontiers of AI to prepare that data throughout each enter and make it accessible by way of any output, in order that it may be actually helpful for you.

That was our imaginative and prescient when we introduced Gemini 1.0 last December. The primary mannequin constructed to be natively multimodal, Gemini 1.0 and 1.5 drove massive advances with multimodality and lengthy context to know data throughout textual content, video, pictures, audio and code, and course of much more of it.

Now tens of millions of builders are constructing with Gemini. And it’s serving to us reimagine all of our merchandise — together with all 7 of them with 2 billion customers — and to create new ones. NotebookLM is a good instance of what multimodality and lengthy context can allow for folks, and why it’s beloved by so many.

During the last 12 months, we’ve been investing in growing extra agentic fashions, which means they will perceive extra in regards to the world round you, suppose a number of steps forward, and take motion in your behalf, together with your supervision.

As we speak we’re excited to launch our subsequent period of fashions constructed for this new agentic period: introducing Gemini 2.0, our most succesful mannequin but. With new advances in multimodality — like native picture and audio output — and native instrument use, it can allow us to construct new AI brokers that carry us nearer to our imaginative and prescient of a common assistant.

We’re getting 2.0 into the arms of builders and trusted testers at this time. And we’re working shortly to get it into our merchandise, main with Gemini and Search. Beginning at this time our Gemini 2.0 Flash experimental mannequin shall be obtainable to all Gemini customers. We’re additionally launching a brand new characteristic referred to as Deep Research, which makes use of superior reasoning and lengthy context capabilities to behave as a analysis assistant, exploring complicated matters and compiling reviews in your behalf. It is obtainable in Gemini Superior at this time.

No product has been remodeled extra by AI than Search. Our AI Overviews now attain 1 billion folks, enabling them to ask totally new sorts of questions — shortly turning into one in all our hottest Search options ever. As a subsequent step, we’re bringing the superior reasoning capabilities of Gemini 2.0 to AI Overviews to sort out extra complicated matters and multi-step questions, together with superior math equations, multimodal queries and coding. We began restricted testing this week and shall be rolling it out extra broadly early subsequent 12 months. And we’ll proceed to carry AI Overviews to extra international locations and languages over the following 12 months.

2.0’s advances are underpinned by decade-long investments in our differentiated full-stack strategy to AI innovation. It’s constructed on customized {hardware} like Trillium, our sixth-generation TPUs. TPUs powered 100% of Gemini 2.0 coaching and inference, and at this time Trillium is generally available to clients to allow them to construct with it too.

If Gemini 1.0 was about organizing and understanding data, Gemini 2.0 is about making it rather more helpful. I can’t wait to see what this subsequent period brings.

-Sundar