Shengshu streams full Motubrain 2.0 embodied model training process

On October 21, 2026, the world model team at Shengshu Technology (生数科技) launched a public livestream covering the full training process of Motubrain 2.0, an embodied model. The broadcast is scheduled to run for half a month and is meant to show the model's progress from pretraining through post-training to capability iteration. The company framed the stream around Physical AI, which it describes as AI that understands physical environments and completes real tasks through robots. Compared with large language models, such systems also have to handle action decisions, task execution and generalization across different scenes.

Motubrain is Shengshu Technology's embodied model series, built to give robots a general-purpose "brain" and to explore how capabilities transfer across tasks, environments and robot bodies. According to the report, Motubrain 2.0 is about to be open-sourced, and the team chose to publish its full training process before the release, showing how the model moves from training and testing to capability iteration.

The public training covers four stages. In pretraining, the team observes how training scale affects model performance and zero-shot generalization. In post-training, the model is trained further on specific tasks to improve task performance. The report stops there and does not detail the remaining stages.

What can be verified ends at the start of the livestream: it began on October 21, is planned to last half a month, and Motubrain 2.0 has not yet been open-sourced. The material does not disclose viewership figures, the compute or data scale used for training, or any concrete evaluation results for the model.