Research
Perceptron AI Ships Isaac 0.5, a 36B Open-Weight Robotics Model Averaging 97.2% on LIBERO
Perceptron AI launched Isaac 0.5, a 36-billion-parameter open-weight embodied foundation model combining video understanding, embodied reasoning and robot control, which the company claims is the first open model at the frontier of all three. On LIBERO it averages 97.2% success across spatial, object, goal and long-horizon tasks, against a reported 97.0% for NVIDIA GR00T N1.7 and 96.9% for pi-0.5 in the same comparison. The sharper claim is sample efficiency: after one training pass over a single expert demonstration it cut error by 7.0x to 10.5x on three unseen tasks where pi-0.5 managed 2.3x to 3.1x. Weights are at huggingface.co/PerceptronAI/Isaac-0.5, created 2026-08-26, with fine-tuning and inference code on GitHub.
↳ Follow the thread