DeepSeek released DeepSeek-R1 under the MIT License, which is on the Open Source Initiative’s approved list; the DeepSeek-V3 weights use the DeepSeek License Agreement, which is not.
5 entriesVersions 2First entry Dec 2024Latest Checked
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software
a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare, publicly display, publicly perform, sublicense, and distribute the Complementary Material, the Model, and Derivatives of the Model.
“Downloadable” means the weights are published for download under the license named in the row. It is not a statement about that license’s terms.
Dated entries
2026
2 entries
Notice: DHS publishes an order excluding DeepSeek products from civilian procurement and requiring removal of DeepSeek products and third-party products developed using DeepSeek implementations.
The notice publishes an order by the Secretary of Homeland Security under the Federal Acquisition Supply Chain Security Act of 2018, 41 U.S.C. 1321-1328. It covers Hangzhou DeepSeek Artificial Intelligence Co., Ltd and affiliated entities as defined in 48 CFR 2.101. The signed order is reproduced as images in the notice’s appendix. The order in the appendix carries an original release date of July 24, 2026; the Federal Register notice publishing it is dated October 20, 2026.
On open weightsIn April 2026, the Center for AI Standards and Innovation (CAISI) evaluated the open-weight AI model DeepSeek V4 Pro ("DeepSeek V4").
The page says CAISI's evaluations, which include non-public benchmarks, indicate that DeepSeek V4 performs similarly to GPT-5, and that DeepSeek's self-reported results placed it nearer to Opus 4.6 and GPT-5.4. It says CAISI served the model from cloud-based H200 and B200 GPUs.
On open weightsTo evaluate DeepSeek models, which are available as open-weight models, CAISI downloaded their model weights from the model sharing platform Hugging Face and deployed the models on CAISI's own cloud-based servers.
The report evaluates DeepSeek R1, R1-0528, and V3.1 against GPT-5, GPT-5-mini, gpt-oss, and Opus 4. It says DeepSeek models were much more likely to follow malicious hijacking instructions than the evaluated U.S. frontier models, and that the evaluated U.S. open-weight model matched or exceeded DeepSeek on that measure.
On open weightsCode and models are released under the MIT License: Distill & commercialize freely!
The same news page says “Fully open-source model & technical report,” “DeepSeek-R1 is now MIT licensed for clear open access,” “Open for the community to leverage model weights & outputs,” and “6 small models fully open-sourced.” The MIT file on the R1 repository begins “Copyright (c) 2023 DeepSeek.” The DeepSeek-V3 LICENSE-MODEL file remained the October 23, 2023 DeepSeek License Agreement when fetched on October 16, 2026.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software
The file on the DeepSeek-V3 repository, fetched October 16, 2026, is still the DeepSeek License Agreement, Version 1.0, October 23, 2023, the same text as on DeepSeek-LLM. It is not the MIT License. The repository lists 163 safetensors shards. Use-based restrictions in Attachment A must be passed downstream, and section 7 reserves a right to restrict usage that violates the license.
a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare, publicly display, publicly perform, sublicense, and distribute the Complementary Material, the Model, and Derivatives of the Model.
Maple runs open-weight models inside a hardware-isolated enclave. Maple cannot see your chats. How it works
Every entry cites one primary source, quotes the document’s own words about open weights, and is checked against that source before it publishes. Corrections keep the old wording visible. Methodology · Corrections · Suggest an entry or a correction
