Commit

2238f8d

It's amazing how far free models can take you. Not talking about local models. Those are really powerful as well, but at a very steep hardware and energy cost. I'm just considering free inference offers "in the cloud". There's a lot they can do. Sure, for complex coding and very deep sessions, probably not. Or consider having a frontier model in the mix as orchestration and validator. But for grunt work, these free models are more than capable of doing 60% of the work. That means we can have about 60% of the frontier tokens available for deeper, more complex work.

commit 2238f8d