DeepSeek R1 671B

DeepSeek R1 671B is the full-scale reasoning model from DeepSeek, trained with reinforcement learning to perform deep chain-of-thought reasoning. Unlike the distilled variants, this is the original teacher model with 671B mixture-of-experts parameters, delivering state-of-the-art reasoning performance. The model excels at complex mathematical proofs, multi-step logical reasoning, and challenging coding problems by explicitly showing its reasoning process. Its MIT license makes it one of the most permissively licensed frontier-class models available.