Config leverages open models as wedge

Config is turning open robot models into a wedge for selling the harder parts of the stack, the messy work of getting usable demonstrations, translating them into robot specific actions, and running them reliably on real machines. That matters because many buyers do not want to bet on one closed model vendor. They want to keep model choice flexible, while still paying for the data pipeline and deployment work that actually makes a robot perform a task on site.1 2 3 4

• Loop lets a customer start with GR00T or Pi 0.5, feed in its own demonstrations, then use Config for collection, conversion, fine tuning, and live deployment. In practice, this means Config can win revenue even when the customer does not buy CFG-1, its in house model.1 4
• Open models also shrink the bundle a buyer must purchase. Instead of buying a closed robot brain plus the tooling around it, a team can use an open model and spend on the pieces that are hardest to build internally, data capture rigs, trajectory conversion for its specific arms and grippers, and deployment feedback loops.1 4 5
• This puts Config in a different lane from Lightwheel and Maniformer. Lightwheel leans more toward infrastructure for customers training their own models, while Maniformer emphasizes datasets, hardware, and a broader supply network. Config sits closer to the execution layer where a customer wants a task working on a real robot, regardless of model choice.2 3 5

The likely direction is that robot models become easier to access, while robot ready data and deployment remain bottlenecks. If that happens, more budget shifts away from proprietary model licenses and toward the companies that can collect demonstrations, adapt them to different robot bodies, and keep policies improving after deployment. That is where Config is positioning itself.1 2 3 4 5