DeepSeek’s First-generation Reasoning Models

DeepSeek’s first-generation thinking designs, attaining efficiency comparable to OpenAI-o1 across mathematics, code, and thinking tasks.

Models

DeepSeek-R1

Distilled models

DeepSeek group has demonstrated that the reasoning patterns of bigger models can be distilled into smaller designs, leading to better efficiency compared to the reasoning patterns discovered through RL on small models.

Below are the models created through fine-tuning against a number of dense designs widely used in the research study community utilizing thinking data generated by DeepSeek-R1. The evaluation results show that the distilled smaller models carry out extremely well on benchmarks.

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 licensed under the MIT License. DeepSeek-R1 series assistance business usage, enable any adjustments and derivative works, including, but not limited to, distillation for training other LLMs.