Smaller models match frontier AI on legal tasks with custom training

Original: Our RL + post training results. Thanks @baseten for collab :

Source: x.com ↗