Toward Large Reasoning Models, DeepSeek-R1, Kimi k1.5 and scaling RL, Mind Evolution for search in planning, improving Process Reward Models in math reasoning.