Sentdex runs a <$1K quadruped on GLM's vision model, testing generic multimodal LLMs as robotics brains

Sentdex · x · 2026-09-04

Sentdex demonstrates an experiment: using a general-purpose multimodal LLM with vision understanding (GLM 5.3 Flash) to handle high-level robotics intelligence, instead of purpose-built VLAs or world models. He argues recent open-source multimodal LLMs have become fast and smart enough to make this viable.

He also recommends a <$1K quadruped from Luwu Dynamics, which he has been testing for 5 years and calls the best one yet; a full video is coming.

Related event: Sentdex drives a quadruped robot with a general multimodal LLM(3 posts)→

Original post →

More from Embodied

Embodied channel →