Efficient Reinforcement Learning for Mathematical Theorem Proving With Compact Language Models. JAAIR 2026, 5 (1).