DeepSeek's reasoning model trained with reinforcement learning. Excels at math, code, and complex reasoning tasks.