DeepSeek launches R1, an open-weight reasoning model under the MIT license

DeepSeek has announced R1, a reasoning model with open weights under the MIT license. According to the company, R1 reaches reasoning capability equivalent to OpenAI o1, accompanied by six smaller distilled models and a low-cost API.

Reasoning models are a class of model that spends extra compute before answering, breaking a problem into intermediate steps instead of responding immediately. For enterprises, the notable point is the MIT license: the weights may be used, modified and deployed internally without a separate permission request. The six distilled models give engineering teams more options when compute infrastructure is limited.

The practical significance lies first in operating cost and control over data. An open model lets enterprises self-host, decide for themselves what data goes in, and reduce dependence on a single vendor. For tasks that need multi-step reasoning, such as analysing long documents or checking contracts for errors, open reasoning models open a deployment path that previously was usually feasible only through closed APIs.

Open weights, however, do not mean easy deployment. Enterprises need to account for infrastructure, electricity and cooling costs when self-operating, for the quality of Vietnamese-language handling compared with commercial models, and for responsibility over output safety. An internal evaluation process on real data should be in place before the model enters business operations, especially with sensitive data.

Source: api-docs.deepseek.com