MiMo-2.6-Pro runs at 0.646 tokens/sec on Jetson AGX Thor

A Reddit user reports loading and running the roughly 566GB FP8 MiMo-2.6-Pro on a 128GB Jetson AGX Thor at only 0.646 tokens/sec, with disk reads during inference. The post jokes about Dario's comments on open-source model transparency versus closed models; it is a personal experiment, not an official release.

• Liquid AI releases open edge decision models2026-10-07
• Reddit post: GLM shots fired at Dario image2026-09-18
• Ask HN: Are The Matrix 'agents' villains by coincidence?2026-09-14