GEN-1

Generalist AI's second-generation embodied foundation model, aiming for reliable, fast “mastery” of simple tasks rather than just getting them done.

GEN-1 is an embodied foundation model released in April 2026 by the US robotics company Generalist AI, the successor to GEN-0. It was pretrained on more than 500,000 hours of real physical-interaction data, mostly collected from people wearing low-cost wearable devices while doing everyday activities. The company defines its goal as “mastery”: reliable, fast, and able to improvise. According to the company, fine-tuning on just about 1 hour of robot data per task brought success on tasks where the previous model scored around 64% up to 99%, folding a paper box in about 12.1 seconds — roughly 3 times faster than the previous best. The August 2026 version, GEN-1.5, gained in-context learning ability (learning from a demonstration without changing its weights): it can perform a new task after watching just a 3-to-12-second demonstration, with an average single-shot success rate of about 59%.

ExampleIn an official demo, GEN-1 folded paper boxes 200 times in a row and assembled blocks 1,800 times in a row, to demonstrate its reliability over long, continuous operation.

Also called

GEN-1.5, Scaling Embodied Foundation Models to Mastery

Related

GEN-0 · Generalist AI · Embodied Foundation Model · In-Context Learning · Scaling Law · Real-Robot Data

Sources

GEN-1: Scaling Embodied Foundation Models to Mastery (Generalist AI Blog)
GEN-1.5: Embodied Foundation Models are One-Shot Learners (Generalist AI Blog)

As of

2026-08