"Doing More with Less Tokens: Hierarchical Reinforcement Learning for Efficient Coding Agents". Official 4B, 9B, and 35B-A3B models.