DeepSeek-R1

DeepSeek-R1 is a reasoning model released by DeepSeek in January 2025, developed through pure reinforcement learning without supervised fine-tuning — a training approach that drew significant attention from the research community for demonstrating that RL alone could produce reasoning capabilities previously associated with closed proprietary systems. Released with open weights and permissive licensing, it shifted community expectations about what openly available reasoning models could achieve, and became a widely cited reference point in discussions about reasoning model training methodology.