The DeepSeek Reasoner (DeepSeek R1) is an advanced open-source AI model distinguished by its reasoning-centric design. Developed by DeepSeek, this model excels in logical inference, mathematical problem-solving, and real-time decision-making. It was trained using a unique approach that combines supervised fine-tuning (SFT) with large-scale reinforcement learning (RL), which enables it to generate sophisticated reasoning chains and self-verify its outputs.

DeepSeek R1 features a Mixture of Experts (MoE) framework, allowing it to specialize in different problem domains while maintaining efficiency. With 671 billion parameters, it achieves performance comparable to top proprietary models like OpenAI's o1, yet it is more cost-efficient and scalable. The model supports multimodal capabilities, and is distributed under the permissive MIT license, making it accessible for commercial and research use cases.

‍

- The deepseek-chat model has been upgraded to DeepSeek-V3. deepseek-reasoner points to the new model DeepSeek-R1.
Source: https://api-docs.deepseek.com/quick_start/pricing