Reasoning models trained on synthetic data using reinforcement learning.