
Lumos model tops global benchmark for zero-shot embodied AI
Despite having only 2.8 billion parameters—less than one-sixth the size of Nvidia Cosmos—Prime R0 delivered the best overall performance.

Despite having only 2.8 billion parameters—less than one-sixth the size of Nvidia Cosmos—Prime R0 delivered the best overall performance.
-600x400.jpg)
Zhu’s comments come as embodied AI and robotics attract growing investment globally, with developers racing to move machines beyond controlled industrial settings and into everyday environments.

On the commercial front, Noematrix’s robots have already been rolled out in pharmacies.

The draft sets ambitious targets: by 2030, Anhui aims to rank among China’s top tier in humanoid robot output and place industrial robot production among the national leaders.

The funding comes as embodied AI developers face a shortage of high-quality training data, inconsistent standards and fragmented datasets.

The models can be deployed independently or jointly, forming a unified base for robotic systems operating in real-world environments.

The startup’s focus is not on predicting the next word but on predicting what happens next in the physical world.

Increasingly, competition is shifting toward capabilities spanning data collection, embodied foundation models, world models and real-world deployment.

The company said the milestone was reached within 10 months of mass production, which it claims is 2-3 years faster than the industry average.

Lumos said participants will not be required to pay licensing fees, provide equity or make upfront purchases.