Over the past decade we have laid many foundations of the contemporary AI era, from pioneering Transformer Architecture on which all gigantic language models are based, to develop agent systems that can learn and plan Alphago AND Alphazero.
We used these techniques to make breaks Quantum calculationIN mathematicsIN Life sciences AND Algorithmic discovery. We will continue to double the width and depth of our basic research, working on inventing the next great breakthroughs necessary for artificial general intelligence (Aga).
That is why we are working on extending our best multimodal foundation model, Gemini 2.5 Pro to become a “world model” that can create plans and imagine recent experiences through understanding and simulating the aspects of the world, like the brain.
We have been making progress in this direction for some time, from our pioneering training agents to master intricate games To go AND Starcraftfor construction GRAND 2which is able to generate simulated 3D environments with which you can interact with a single picture line.
We already see evidence of these possibilities appearing in Gemini’s ability to employ global knowledge and reasoning to represent and simulate natural environmentsIN I seedeep understanding of intuitive physics and path Gemini robotics Teaches work to capture, follow the instructions and adapt in flight.
Making Gemini a global model is a key step in developing a recent, more general and more useful type of AI – a universal AI assistant. This is clever artificial intelligence, understands the context in which you are, which can plan and take action on your behalf, on every device.