Short answer
For DeepSeek-R1, the clearest short answer is: it is open-weight, and DeepSeek says its R1 code repository and model weights are MIT-licensed. Those facts do not, by themselves, establish that the release meets the Open Source Initiative’s (OSI) Open Source AI Definition.
OSI’s definition calls for the freedoms to use, study, modify, and share an AI system, along with access to the preferred form for making changes. For machine-learning systems, that form includes model parameters, complete source code used to train and run the system, and sufficiently detailed information about training data. (OSI definition)
OSI’s January 30, 2025 assessment said DeepSeek had shared model weights and architecture, but not code or training and evaluation data openly; it concluded DeepSeek did not meet the Open Source AI Definition. That is the assessment published at that time, not a finding that an exact rebuild or release of every original data file is required. (OSI’s January 30, 2025 assessment)
Open weight vs. open source
Open-weight describes a release whose trained model parameters are available. That can let users run or adapt the model, subject to the applicable license and practical requirements. It does not tell you, on its own, whether the materials needed to study and modify the system under a broader definition are also available.
Under OSI’s definition, training-data requirements do not necessarily mean publishing every original data file. The definition calls for detailed data information sufficient for a skilled person to build a substantially equivalent system, including descriptions of data that cannot be shared and where available data can be obtained. It also calls for the complete source code used to train and run the system. (OSI definition; OSI FAQ)
So, if you mean “Are DeepSeek-R1 weights available?” the answer is yes. If you mean “Does R1 meet OSI’s Open Source AI Definition?” OSI’s published January 2025 assessment said no, citing code and data materials it said were not openly shared. Those are different questions; don’t treat “open weights” or an MIT license as proof of the broader classification.
What license applies to DeepSeek-R1?
DeepSeek’s R1 repository states that its code repository and model weights are licensed under the MIT License and that the R1 series supports commercial use, modifications, and derivative works. The license text requires retaining the copyright and permission notice in copies or substantial portions of the software. Check the license and model card for the exact artifact and version you plan to use, particularly if you are selecting a specific distilled release.
A license sets permissions and conditions for covered material. It does not establish that the release includes the code and data information called for by OSI’s broader AI definition. “MIT-licensed open weights” is therefore more precise than using “open source” as a blanket description without specifying the standard.
Open weights or a production API?
These are different ways to access a model:
| Route | What you use | Main trade-off |
|---|---|---|
| Download the weights | Model files you deploy in an environment you choose | More control over deployment, but you operate and secure the infrastructure |
| Use a hosted API | A provider-operated service that accepts requests and returns model responses | Less model infrastructure to manage, but you depend on the provider’s service, terms, and available models |
DeepSeek documents hosted API access separately from downloadable model weights. An API call is not the same as downloading or self-hosting R1 weights, and API model offerings may differ from released weight names. Check the current API model list and service terms when evaluating a production integration.
A team that needs to run inference in its own environment might evaluate downloadable weights and account for the hardware, deployment, and security work. A team that does not want to operate model infrastructure might evaluate a hosted API and review the provider’s current terms. Neither route is automatically private or suitable for production just because of its label.
Bottom line
Describe DeepSeek-R1 as open-weight and MIT-licensed; don’t treat those facts alone as proof that it meets OSI’s Open Source AI Definition. OSI’s January 30, 2025 assessment said R1 did not meet that definition, citing code and training/evaluation data that it said were not openly shared. The definition asks for sufficiently detailed training-data information—not necessarily every original data file—and complete training and running code. For a deployment decision, check the license for the exact model artifact, or review current API documentation and service terms if you plan to use the hosted route.
Sources
The Open Source AI Definition – 1.0