Skip to content

DeepSeek-R1

An open reasoning model trained with large-scale reinforcement learning

Visit official site

What DeepSeek-R1 is

The model is described as the first-generation reasoning model from the DeepSeek team, trained with large-scale reinforcement learning without supervised fine-tuning. The material states it performs on mathematics, code and reasoning tasks at a level comparable to OpenAI-o1.

What you can do with it

  • Work on mathematics problems, code generation and natural language understanding
  • Choose a distilled variant sized for different needs
  • Use open code and model weights for research
  • Apply commercial use and secondary development including distillation

Getting started

  • Visit the DeepSeek-R1 GitHub page
  • Pick the main model or one of the distilled versions
  • Launch the model service with tools such as vLLM or SGLang
  • Configure parameters such as temperature and context length

Model notes

  • Distilled models are built on the Llama and Qwen series
  • Training relies on large-scale reinforcement learning rather than supervised fine-tuning
  • Open code and weights are provided for research and development
  • The material states commercial use and modification are supported

Pros & cons

✓ What we like

  • Mathematics, code generation and natural language understanding are named
  • Modification and secondary development including distillation are allowed

! What to watch out for

  • Service startup is described through external tools such as vLLM or SGLang
  • Reasoning quality depends on configured parameters such as temperature and context length

FAQ

How was the model trained?

The material states large-scale reinforcement learning without supervised fine-tuning.

Are the weights available?

The material lists open code and model weights, with commercial use and secondary development allowed.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review