History
DeepSeek-R1 opens a reasoning model for reuse
DeepSeek released R1 under an MIT licence with its own comparisons against OpenAI o1.

On 20 January 2025, DeepSeek released its R1 reasoning model. The company reported performance comparable to OpenAI o1 on mathematics, coding and reasoning tasks. These are the maker’s claims. DeepSeek published code and models under the MIT licence. The release page describes the launch.

With R1, access to the model and its usage rights deserve attention together. The discussion also concerned what you could build and change with a model. A comparison from its maker does not replace a trial using your own work.

For a small or medium-sized business, one possible trial is checking an order against clear delivery conditions. Give the same assignment to different models. Define beforehand which conditions must never be broken and how you will check the answer. For a managed installation, also ask who maintains the chosen version and which components it includes. Assess quality, usage rights and maintenance separately. Available model weights are a starting point for an application; you still need to demonstrate the step towards a useful business process.