Steiner

A series of open-source reasoning models trained with reinforcement learning on synthetic data. They explore several reasoning paths at inference and verify or backtrack on their own.