Was about to object, then realized you linked Allen ai and had the useful caveat. For what it's worth the OpenMDW license, which Nemotron and a few others have

Was about to object, then realized you linked Allen ai and had the useful caveat. For what it's worth the OpenMDW license, which Nemotron and a few others have adopted, does say model weight. That said, I've noticed the training procedures and corpus size of more useful open/available weights models are settling down more than I expected. Wonder if crowd sourcing good training data, even if it's just expensive model coding session transcripts, has potential to level the landscape some.