Some DeepSeek releases are described as open source or have released model weights, but there is no single license that applies to every DeepSeek model. DeepSeek-R1’s README says its code and weights are MIT-licensed; DeepSeek-V3 has an MIT-licensed code repository but a separate license for model use; and the DeepSeek Coder project page calls Coder open source. Check the exact model and variant before relying on those terms. (DeepSeek-R1 · DeepSeek-V3 · DeepSeek Coder)
“Open source” can mean different things in AI. Code, model weights, training data, and the full training process are separate components; a release or license for one does not establish that all the others are public. DeepSeek’s V3 model license, for example, says the training data is not licensed under that agreement. (V3 Model License)
What the licenses say about V3, R1, and Coder
| Model or release | What the official documentation says | What to check |
|---|---|---|
| DeepSeek-V3 | The repository code is MIT-licensed. Use of V3 Base/Chat models is subject to a separate Model License. The README says the V3 series supports commercial use. | Read the Model License, not just the repository’s MIT license. It includes use-based restrictions, so the commercial-use statement should not be read as permission for every possible use. (V3 repository · V3 Model License) |
| DeepSeek-R1 | The README says the code repository and model weights are MIT-licensed, and that the series allows commercial use, modifications, and derivative works. | Check the exact checkpoint and applicable upstream terms. The README identifies R1-Distill-Qwen-1.5B, -7B, -14B, and -32B as derived from Qwen-2.5 (Apache 2.0); R1-Distill-Llama-8B as derived from Llama 3.1; and R1-Distill-Llama-70B as derived from Llama 3.3. (R1 README) |
| DeepSeek Coder | Its project page calls the Coder series “open source and free for research and commercial use.” | The page does not name a specific license. Don’t infer a license name or apply the statement to other DeepSeek families. (DeepSeek Coder) |
These are examples, not a blanket license statement for the whole DeepSeek catalog. For a commercial project, identify the exact model and variant, read its official license and model documentation, and check any upstream terms that apply. These summaries are not a legal review of a particular deployment.
Do open weights mean the model is available through a production API?
No. Downloadable weights and a hosted API are separate ways to use a model. A weight release does not by itself establish that a particular provider offers that model as an API or specify the provider’s model identifier, limits, pricing, or service terms.
DeepSeek’s API documentation provides a model-list endpoint for checking the models and capabilities available through its hosted service. Its API documentation identifies deepseek-v4-pro and deepseek-flash as model names. It says deepseek-flash is served by DeepSeek-V4.1-Flash-1017; the model name remains deepseek-flash. These API identifiers describe hosted access, not the license or self-hosting terms for the underlying weights. (Model-list endpoint · API documentation)
Before building a production integration, check the provider’s current documentation for the exact API model ID and supported interface, then review applicable limits, pricing, data-handling, and service terms. If you plan to run weights yourself, check the model’s license and deployment requirements instead. Don’t treat API availability as proof that a model is open source—or model openness as proof that an API is available.
Practical takeaway
The most accurate short answer is: some DeepSeek releases are described as open source or have openly available weights, but the answer depends on the specific model and on what you mean by “open.” V3, R1, and Coder have different documentation and license details. For a production choice, verify two things independently: the exact model’s license for your intended use, and the terms of the hosted API or self-hosted deployment you plan to use.
Sources
DeepSeek-V3 — official GitHub repository
DeepSeek-R1 — official GitHub repository