Bagel Labs

Bagel Labs is a physical AI research lab based in San Francisco and Toronto. It has developed an architecture that uses dramatically less data and compute to train while delivering a state-of-the-art model-size/performance tradeoff. WorldDiT reaches the parameter-success Pareto frontier across all four LIBERO simulation suites. Paris 1.0 improves FID by 24 percent over a monolithic baseline at matched aggregate compute, paving the way for Paris 2.0. Paris 2.0 extends the decentralized training architecture to video, cutting FVD roughly in half using the same data and matched total compute and paving the way for physical AI world models.

Bagel Labs researchers have authored peer-reviewed papers in the main conference tracks at NeurIPS, ICML, ICLR, CVPR, and ECCV.