The numbers on this new implementation from MIT are wild. It manages to be feature-complete while running on ultra-low-end hardware. It’s significantly more efficient than both OpenClaw and PicoClaw. It’s great to see this kind of extreme optimization coming out of the US research scene, really pushing the limits of what $10 hardware can do. What do you guys think of the performance gains?r