Browse Models
DeepSeek R1 is a family of language models developed by Deepseek AI that specializes in advanced reasoning and chain-of-thought problem-solving. Released in January 2025, the family includes the flagship DeepSeek-R1 model with 671 billion parameters using Mixture-of-Experts architecture, along with smaller distilled variants based on Qwen and Llama architectures ranging from 1.5B to 70B parameters, all trained using reinforcement learning techniques and knowledge distillation.