And finally, and perhaps most importantly, on January 20, DeepSeek rolled out its DeepSeek-R1 model, which adds two more reinforcement learning stages and two supervised fine tuning stages to enhance ...