[christkv]

I guess we are going to be using multiple small specialized models with a reasoning model and tooling.

[selim-now]

isn't that the premise of the Nvidia paper? https://arxiv.org/pdf/2506.02153