Introducing GEN-1

We're developing generalist intelligence from the physical world. And today we're introducing our most advanced model, Gen One. It's a brain for many robots. It really understands the physical world and how to act in it. It's trained from scratch on our data set of half a million hours of physical experience. We believe it's the first model to master a broad range of physical skills. It can enable robots to do servicing of other robots. It can fold laundry. It can do industrial kitting of automotive parts. But the point isn't that it can do any one of those tasks. The point is that for the first time in robot learning, we can start to do many of these tasks and do them very well and learn how to do them in a very short amount of time. Gen one can also learn from experience how to complete tasks faster over time, up to three times faster than the state-of-the-art. And it's showing these levels of improvisational intelligence that we've never seen before. It's kind of like for those that remember opening up chat GBT for the first time and asking it to write a poem about something that clearly nobody had ever written a poem about before, like write a haik coup on who's going to win the 2026 NBA title. And it's that ability to connect ideas from different places in order to solve new problems. That's really what we're starting to see emerge from these models. Gen one builds on the foundation of Gen Zero, which we released a handful of months ago. And Gen Zero really brought scaling laws to robotics for the first time, showing that with more compute and data, you get predictable advances in generalization. And now with Gen One, we've both scaled quite a bit further and accelerated with some innovations across the stack. And these models are really starting to be able to master tasks. When we think about mastery, we think about reliability and speed and improvisation. And reliability and speed, these have been around in robotics for decades. We've had industrial robots deployed on automanufacturing lines since the early 1960s. But it's really about combining both of those two with this third capability of improvisation. It's a particular type of open-ended problem solving that we humans learn to do in the physical world before the digital one. For us, Gen One is more than just a model. It captures a very real and we think important part of artificial intelligence that's missing from the chatbots that we have today. It's the opened problem solving set of skills born from acting in a messy world. It's about adapting to change and fixing problems as they come. And for machines, it's only through experiencing the physical world that all the text on Wikipedia can finally start to make sense. And I think that we are still in the very early days and this is just one step on the journey.

We've created GEN-1, our latest milestone in scaling robot learning. We believe it to be the first general-purpose AI model that crosses a new performance threshold: mastery of simple physical tasks. It improves average success rates to 99% on tasks where previous models achieve 64%, completes tasks roughly 3x faster than state of the art, and requires only 1 hour of robot data for each of these results. GEN-1 unlocks commercial viability across a broad range of applications—and while it cannot solve all tasks today, it is a significant step towards our mission of creating generalist intelligence for the physical world. For more see https://generalistai.com/blog/apr-02-2026-GEN-1