Generalist AI has released GEN-1, an embodied foundation model for robotics that achieves 99% success rates on real-world tasks — a massive leap from the 64% achieved by its predecessor. The model completes tasks up to three times faster and requires only about one hour of robot-specific data to adapt to new tasks, making it one of the most data-efficient robotics AI systems to date.
GEN-1 was trained from scratch on half a million hours of real-world human activity data collected through low-cost wearable devices, bypassing the need for expensive teleoperation or simulation datasets. In demonstrations, robots running GEN-1 folded t-shirts 86 times consecutively, serviced robot vacuums over 200 times, packed blocks more than 1,800 times, and folded boxes over 200 times — all with minimal errors.
The company describes GEN-1's key advantage as "intelligent improvisation" — the ability to adapt to unexpected situations in unstructured environments. Early access is now available to selected partners as Generalist AI continues developing the platform.
