DeepSeek's first-generation thinking models, achieving performance comparable to OpenAI-o1 across mathematics, code, and thinking tasks.

Models

DeepSeek-R1

Distilled designs

DeepSeek team has actually shown that the thinking patterns of bigger models can be distilled into smaller designs, leading to better efficiency compared to the thinking patterns discovered through RL on little models.

Below are the designs created by means of fine-tuning versus several dense designs widely used in the research community using thinking information created by DeepSeek-R1. The evaluation results demonstrate that the distilled smaller thick models perform extremely well on criteria.

DeepSeek-R1-Distill-Qwen-1.5 B

DeepSeek-R1-Distill-Qwen-7B

DeepSeek-R1-Distill-Llama-8B

DeepSeek-R1-Distill-Qwen-14B

DeepSeek-R1-Distill-Qwen-32B

DeepSeek-R1-Distill-Llama-70B

License

The design weights are accredited under the MIT License. DeepSeek-R1 series support commercial use, enable for any modifications and acquired works, including, however not limited to, distillation for training other LLMs.