# Open-source artificial intelligence
> Mediated Wiki article. Canonical URL: https://mediated.wiki/source/Open-source_artificial_intelligence
> Markdown URL: https://mediated.wiki/source/Open-source_artificial_intelligence.md
> Source: https://en.wikipedia.org/wiki/Open-source_artificial_intelligence
> Source revision: 1350744847
> License: Creative Commons Attribution-ShareAlike 4.0 International (https://creativecommons.org/licenses/by-sa/4.0/)
{{Short description|Concept of open-source software applied to AI}}
{{cs1 config|display-authors=6}}
{{Artificial intelligence}}
{{Use dmy dates|date=January 2026}}
'''Open-source artificial intelligence''', as defined by the [Open Source Initiative](/source/Open_Source_Initiative), is an AI system that is freely available to use, study, modify, and share.[{{cite news |last1=Williams |first1=Rhiannon |last2=O'Donnell |first2=James |date=22 August 2024 |title=We finally have a definition for open-source AI |url=https://www.technologyreview.com/2024/08/22/1097224/we-finally-have-a-definition-for-open-source-ai/ |access-date=28 November 2024 |work=MIT Technology Review |language=en}}][{{cite news |last1=Robison |first1=Kylie |date=28 October 2024 |title=Open-source AI must reveal its training data, per new OSI definition |url=https://www.theverge.com/2024/10/28/24281820/open-source-initiative-definition-artificial-intelligence-meta-llama |access-date=28 November 2024 |work=The Verge |language=en}}] This includes [datasets used to train the model](/source/Training%2C_validation%2C_and_test_data_sets), its code, and [model parameters](/source/Model_parameter), promoting a collaborative and transparent approach to AI development so someone could create a substantially similar result.[{{Cite web |title=The Open Source AI Definition – 1.0 |url=https://opensource.org/ai/open-source-ai-definition |url-status=live |archive-url=https://web.archive.org/web/20250331184828/https://opensource.org/ai/open-source-ai-definition |archive-date=31 March 2025 |access-date=14 November 2024 |website=[Open Source Initiative](/source/Open_Source_Initiative) |language=en-US}}][{{Cite news |date=6 November 2024 |title=A battle is raging over the definition of open-source AI |url=https://www.economist.com/science-and-technology/2024/11/06/a-battle-is-raging-over-the-definition-of-open-source-ai |access-date=9 December 2025 |newspaper=The Economist |issn=0013-0613}}]
The debate over what should count as ‘open-source’ given a range of openness among AI projects has been significant. Some large language models touted as open-sourced that only release model-weights (but not training data and code)[{{Cite web |title=Open Weights: not quite what you've been told |url=https://opensource.org/ai/open-weights/ |access-date=23 September 2025 |website=Open Source Initiative |language=en-US}}][{{Cite web |last1=Capoot |first1=Ashley |last2=Sigalos |first2=MacKenzie |date=5 August 2025 |title=OpenAI releases lower-cost models to rival Meta, Mistral and DeepSeek |url=https://www.cnbc.com/2025/08/05/openai-open-weight-meta-mistral-deepseek-ai.html |access-date=23 September 2025 |website=CNBC |language=en}}] have been criticized as "[openwashing](/source/openwashing)"[{{cite book |last1=Liesenfeld |first1=Andreas |title=The 2024 ACM Conference on Fairness, Accountability, and Transparency |last2=Dingemanse |first2=Mark |date=5 June 2024 |publisher=Association for Computing Machinery |isbn=979-8-4007-0450-5 |pages=1774–1787 |chapter=Rethinking open source generative AI: Open washing and the EU AI Act |doi=10.1145/3630106.3659005 |doi-access=free}}] systems that are mostly closed.[{{cite journal |last1=Castelvecchi |first1=Davide |date=29 June 2023 |title=Open-source AI chatbots are booming — what does this mean for researchers? |journal=[Nature](/source/Nature_(journal)) |volume=618 |issue=7967 |pages=891–892 |bibcode=2023Natur.618..891C |doi=10.1038/d41586-023-01970-6 |pmid=37340135}}] Debate over the benefits and risks of open-sourced AI involve a range of factors like security, privacy and technological advancement.[{{cite journal |last1=Toma |first1=Augustin |last2=Senkaiahliyan |first2=Senthujan |last3=Lawler |first3=Patrick R. |last4=Rubin |first4=Barry |last5=Wang |first5=Bo |date=December 2023 |title=Generative AI could revolutionize health care — but not if control is ceded to big tech |journal=Nature |language=en |volume=624 |issue=7990 |pages=36–38 |bibcode=2023Natur.624...36T |doi=10.1038/d41586-023-03803-y |pmid=38036861}}][{{cite journal |last1=Widder |first1=David Gray |last2=Whittaker |first2=Meredith |last3=West |first3=Sarah Myers |date=November 2024 |title=Why 'open' AI systems are actually closed, and why this matters |journal=Nature |language=en |volume=635 |issue=8040 |pages=827–833 |bibcode=2024Natur.635..827W |doi=10.1038/s41586-024-08141-1 |issn=1476-4687 |pmid=39604616 |doi-access=free}}][{{cite news |last=Davies |first=Pascale |date=20 February 2024 |title=What is open source AI and why is profit so important to the debate? |url=https://www.euronews.com/next/2024/02/20/open-source-vs-closed-source-ai-whats-the-difference-and-why-does-it-matter |access-date=28 November 2024 |work=[Euronews](/source/Euronews) |language=en}}]
== History ==
{{See also|History of artificial intelligence|Timeline of artificial intelligence}}
The history of open-source artificial intelligence is intertwined with both the development of AI technologies and the growth of the [open-source software movement](/source/open-source_software_movement).[{{Cite web |last=Morrone |first=Megan |date=15 February 2024 |title=With the rise of AI, the software business redefines "open" |url=https://www.axios.com/2024/02/15/open-source-ai-definition-openai-meta |access-date=16 December 2025 |website=Axios |language=en}}]
=== 1990s: Early development of AI and open-source software ===
The concept of AI dates back to the mid-20th century, when computer scientists like [Alan Turing](/source/Alan_Turing) and [John McCarthy](/source/John_McCarthy_(computer_scientist)) laid the groundwork for modern AI theories and algorithms.[{{Cite web |title=Appendix I: A Short History of AI {{!}} One Hundred Year Study on Artificial Intelligence (AI100) |url=https://ai100.stanford.edu/2016-report/appendix-i-short-history-ai?utm_source=chatgpt.com |access-date=24 November 2024 |website=ai100.stanford.edu |language=en}}] An early form of AI, the [natural language processing](/source/natural_language_processing) "doctor" [ELIZA](/source/ELIZA), was re-implemented and shared in 1977 by Jeff Shrager as a BASIC program, and soon translated to many other languages. Early AI research focused on developing [symbolic reasoning systems](/source/Symbolic_artificial_intelligence) and [rule-based expert systems](/source/Rule-based_system).[{{Cite journal |last=Kautz |first=Henry |date=31 March 2022 |title=The Third AI Summer: AAAI Robert S. Engelmore Memorial Lecture |url=https://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/19122 |journal=[AI Magazine](/source/AI_Magazine) |language=en |volume=43 |issue=1 |pages=105–125 |doi=10.1002/aaai.12036 |issn=2371-9621}}]{{Better source needed|reason=The current source is insufficiently reliable ([WP:NOTRS](/source/WP%3ANOTRS)).|date=January 2026}}
During this period, the idea of open-source software was beginning to take shape, with pioneers like [Richard Stallman](/source/Richard_Stallman) advocating for free software as a means to promote collaboration and innovation in programming.[{{Cite web |last=Stallman |first=Richard |title=Why Software Should Be Free - GNU Project - Free Software Foundation |url=https://www.gnu.org/philosophy/shouldbefree.en.html |url-status=live |archive-url=https://web.archive.org/web/20241201200252/https://www.gnu.org/philosophy/shouldbefree.en.html |archive-date=1 December 2024 |access-date=24 November 2024 |website=www.gnu.org}}]{{Primary source inline|date=January 2026}} The [Free Software Foundation](/source/Free_Software_Foundation), founded in 1985 by Stallman, was one of the first major organizations to promote the idea of software that could be freely used, modified, and distributed. The ideas from this movement eventually influenced the development of open-source AI, as more developers began to see the potential benefits of open collaboration in software creation, including AI models and algorithms.[{{Cite web |title=The Power of Collaboration: How Open-Source Projects are Advancing AI |url=https://www.kdnuggets.com/2023/08/power-collaboration-opensource-projects-advancing-ai.html |website=kdnuggets.com}}]{{Better source needed|reason=The current source is insufficiently reliable ([WP:NOTRS](/source/WP%3ANOTRS)).|date=December 2025}}[{{Cite web |last=Daigle |first=Kyle |date=8 November 2023 |title=Octoverse: The state of open source and rise of AI in 2023 |url=https://github.blog/news-insights/research/the-state-of-open-source-and-ai/ |access-date=24 November 2024 |website=The GitHub Blog |language=en-US}}]{{Better source needed|reason=The current source is insufficiently reliable ([WP:NOTRS](/source/WP%3ANOTRS)).|date=December 2025}}
In the 1990s, open-source software began to gain more traction,[{{Cite web |last=Code |first=Linux |date=3 November 2024 |title=A Brief History of Open Source |url=https://thelinuxcode.com/a-brief-history-of-open-source/ |access-date=24 November 2024 |website=TheLinuxCode |language=en-US |archive-date=3 December 2024 |archive-url=https://web.archive.org/web/20241203024236/https://thelinuxcode.com/a-brief-history-of-open-source/ |url-status=dead }}]{{Better source needed|reason=The current source is insufficiently reliable ([WP:NOTRS](/source/WP%3ANOTRS)).|date=December 2025}} the rise of machine learning and statistical methods also led to the development of more practical AI tools. In 1993, the CMU Artificial Intelligence Repository was initiated, with a variety of openly shared software.[{{Cite web |title=Topic: (/) |url=https://www.cs.cmu.edu/Groups/AI/0.html |access-date=11 September 2025 |website=www.cs.cmu.edu}}]{{Better source needed|reason=The current source is insufficiently reliable ([WP:NOTRS](/source/WP%3ANOTRS)).|date=December 2025}}
=== 2000s: Emergence of open-source AI ===
In the early 2000s open-source AI began to take off, with the release of more user-friendly foundational libraries and frameworks that were available for anyone to use and contribute to.[{{Cite web |last=Priya |date=28 March 2024 |title=The Evolution of Open Source AI Libraries: From Basement Brawls to AI All-Stars |url=https://www.thegen.ai/post/the-evolution-of-open-source-ai-libraries-from-basement-brawls-to-ai-all-stars |access-date=24 November 2024 |website=TheGen.AI |language=en}}]{{Better source needed|reason=The current source is insufficiently reliable ([WP:NOTRS](/source/WP%3ANOTRS)).|date=December 2025}}
[OpenCV](/source/OpenCV) was released in 2000[{{cite journal |last1=Pulli |first1=Kari |last2=Baksheev |first2=Anatoly |last3=Kornyakov |first3=Kirill |last4=Eruhimov |first4=Victor |date=1 April 2012 |title=Realtime Computer Vision with OpenCV |journal=ACM Queue |volume=10 |issue=4 |pages=40:40–40:56 |doi=10.1145/2181796.2206309 |doi-access=free}}] with a variety of traditional AI algorithms like [decision trees](/source/Decision_tree_learning), [k-Nearest Neighbors](/source/K-nearest_neighbors_algorithm) (kNN), [Naive Bayes](/source/Naive_Bayes_classifier) and [Support Vector Machines](/source/Support_vector_machine) (SVM).[{{cite book |author1=Adrian Kaehler |url=https://books.google.com/books?id=SKy3DQAAQBAJ&pg=PT26 |title=Learning OpenCV 3: Computer Vision in C++ with the OpenCV Library |author2=Gary Bradski |date=14 December 2016 |publisher=O'Reilly Media |isbn=978-1-4919-3800-3 |pages=26ff}}]
=== 2010s: Rise of open-source AI frameworks ===
Open-source deep learning framework as [Torch](/source/Torch_(machine_learning)) was released in 2002 and made open-source with Torch7 in 2011, and was later augmented by [PyTorch](/source/PyTorch), and [TensorFlow](/source/TensorFlow).[{{Cite journal |last1=Costa |first1=Carlos J. |last2=Aparicio |first2=Manuela |last3=Aparicio |first3=Sofia |last4=Aparicio |first4=Joao Tiago |date=January 2024 |title=The Democratization of Artificial Intelligence: Theoretical Framework |journal=[Applied Sciences](/source/Applied_Sciences_(journal)) |language=en |volume=14 |issue=18 |page=8236 |doi=10.3390/app14188236 |hdl=10362/173131 |issn=2076-3417 |doi-access=free |hdl-access=free}}]
[AlexNet](/source/AlexNet) was released in 2012.[{{Cite web |last=Lee |first=Timothy B. |date=11 November 2024 |title=How a stubborn computer scientist accidentally launched the deep learning boom |url=https://arstechnica.com/ai/2024/11/how-a-stubborn-computer-scientist-accidentally-launched-the-deep-learning-boom/ |access-date=11 September 2025 |website=Ars Technica |language=en}}]
[OpenAI](/source/OpenAI) was founded in 2015 with a mission to create open-source artificial intelligence that benefited humanity, at least in part to help with recruitment in the early phases of the organization.[{{Cite news |last=Metz |first=Rachel |date=15 March 2024 |title=OpenAI and the Fierce AI Industry Debate Over Open Source |url=https://www.bloomberg.com/news/newsletters/2024-03-15/openai-tumult-raises-question-of-how-open-an-ai-company-should-be |access-date=16 December 2025 |work=Bloomberg.com |language=en}}] [GPT-1](/source/GPT-1) was released in 2018.
=== 2020s: Open-weight and open-source generative AI ===
With the announcement of [GPT-2](/source/GPT-2) in 2019, OpenAI originally planned to keep the source code of their models private citing concerns about malicious applications.[{{Cite web |last=Xiang |first=Chloe |date=28 February 2023 |title=OpenAI Is Now Everything It Promised Not to Be: Corporate, Closed-Source, and For-Profit |url=https://www.vice.com/en/article/openai-is-now-everything-it-promised-not-to-be-corporate-closed-source-and-for-profit/ |access-date=14 November 2024 |website=VICE |language=en-US}}] After OpenAI faced public backlash, however, it released the source code for GPT-2 to GitHub three months after its release.[{{Cite web |last=Hao |first=Karen |date=23 September 2020 |title=OpenAI is giving Microsoft exclusive access to its GPT-3 language model |url=https://www.technologyreview.com/2020/09/23/1008729/openai-is-giving-microsoft-exclusive-access-to-its-gpt-3-language-model/ |url-status=live |archive-url=https://web.archive.org/web/20210205121656/https://www.technologyreview.com/2020/09/23/1008729/openai-is-giving-microsoft-exclusive-access-to-its-gpt-3-language-model/ |archive-date=5 February 2021 |access-date=8 December 2024 |website=MIT Technology Review |language=en}}] At the time of GPT-3's release GPT-2 was still the most powerful open source language model in the world. 2022 also saw the rise of larger and more powerful models under licenses of varying openness including Meta's OPT.[{{cite web |last=Heaven |first=Will |date=3 May 2022 |title=Meta has built a massive new language AI—and it's giving it away for free |url=https://www.technologyreview.com/2022/05/03/1051691/meta-ai-large-language-model-gpt3-ethics-huggingface-transparency/ |access-date=26 December 2023 |website=[MIT Technology Review](/source/MIT_Technology_Review)}}]
The [Open Source Initiative](/source/Open_Source_Initiative) consulted experts over two years to create a definition of "open-source" that would fit the needs of AI software and models. The most controversial aspect relates to data access, since some models are trained on sensitive data which can't be released. In 2024, they published the Open Source AI Definition 1.0 (OSAID 1.0).[{{Cite web |last=Nicol-Schwarz |first=Kai |date=2 December 2025 |title=French AI lab Mistral releases new AI models as it looks to keep pace with OpenAI and Google |url=https://www.cnbc.com/2025/12/02/mistral-unveils-new-ai-models-in-bid-to-compete-with-openai-google.html |access-date=5 December 2025 |website=CNBC |language=en}}][{{Cite web |last=Heikkilä |first=Melissa |date=2 December 2025 |title=Mistral unveils new models in race to gain edge in 'open' AI |url=https://www.ft.com/content/bc9339a6-a8e4-4c6d-b77a-f9cffafb8a9f |access-date=5 December 2025 |website=[Financial Times](/source/Financial_Times)}}] along with [MosaicML](/source/MosaicML)'s smaller open-source models.[{{Cite news |last=Quach |first=Katyanna |date=22 June 2023 |title=Bigger not always better in AI, boutique models are coming |url=https://www.theregister.com/2023/06/22/small_custom_ai_models/ |archive-url=https://web.archive.org/web/20251214180156/https://www.theregister.com/2023/06/22/small_custom_ai_models/ |archive-date=14 December 2025 |access-date=28 January 2026 |work=[The Register](/source/The_Register) |language=en |url-status=live }}][{{Cite news |last=Quach |first=Katyanna |date=27 June 2023 |title=Databricks snaps up MosaicML to build private AI models |url=https://www.theregister.com/2023/06/27/databricks_mosaicml_merger/ |archive-url=https://web.archive.org/web/20251110012849/https://www.theregister.com/2023/06/27/databricks_mosaicml_merger/ |archive-date=10 November 2025 |access-date=28 January 2026 |work=[The Register](/source/The_Register) |language=en |url-status=live }}] The release of the Llama models was a milestone in generating interest in open-weight and open-source models.[{{Cite journal |last1=Vake |first1=Domen |last2=Šinik |first2=Bogdan |last3=Vičič |first3=Jernej |last4=Tošić |first4=Aleksandar |date=5 March 2025 |title=Is Open Source the Future of AI? A Data-Driven Approach |journal=Applied Sciences |language=en |volume=15 |issue=5 |pages=9–10 |doi=10.3390/app15052790 |doi-access=free |issn=2076-3417}}] In 2024, Meta released a collection of large AI models, including [Llama](/source/Llama_(language_model)) 3.1 405B, which was competitive with less open models.[{{Cite web |last=Mirjalili |first=Seyedali |date=1 August 2024 |title=Meta just launched the largest 'open' AI model in history. Here's why it matters |url=https://theconversation.com/meta-just-launched-the-largest-open-ai-model-in-history-heres-why-it-matters-235689 |access-date=14 November 2024 |website=The Conversation |language=en-US}}] Meta's description of Llama as open-source has been disputed due to Llama's software license, which prohibits it from being used for some purposes, and due to Meta not disclosing the origin of the data used to train the models.[{{Cite news |last=Waters |first=Richard |date=17 October 2024 |title=Meta under fire for 'polluting' open-source |url=https://www.ft.com/content/397c50d8-8796-4042-a814-0ac2c068361f |access-date=14 November 2024 |work=[Financial Times](/source/Financial_Times)}}][{{cite web |last1=Edwards |first1=Benj |date=18 July 2023 |title=Meta launches Llama 2, a source-available AI model that allows commercial applications |url=https://arstechnica.com/information-technology/2023/07/meta-launches-llama-2-an-open-source-ai-model-that-allows-commercial-applications/ |url-status=live |archive-url=https://web.archive.org/web/20231107082612/https://arstechnica.com/information-technology/2023/07/meta-launches-llama-2-an-open-source-ai-model-that-allows-commercial-applications/ |archive-date=7 November 2023 |access-date=14 December 2024 |website=[Ars Technica](/source/Ars_Technica)}}][{{cite web |last=Thomas |first=Prasanth Aby |date=5 November 2024 |title=Meta offers Llama AI to US government for national security |url=https://www.cio.com/article/3599448/meta-offers-llama-ai-to-us-government-for-national-security.html |url-status=live |archive-url=https://web.archive.org/web/20241214234902/https://www.cio.com/article/3599448/meta-offers-llama-ai-to-us-government-for-national-security.html |archive-date=14 December 2024 |access-date=14 December 2024 |website=[CIO](/source/CIO_(magazine))}}]
[DeepSeek](/source/DeepSeek) released their V3 LLM in December 2024, and their R1 [reasoning model](/source/Model-based_reasoning) on 20 January 2025, both as open-weights models under the MIT license.[{{cite web |last=Chen |first=Caiwei |date=24 January 2025 |title=How a top Chinese AI model overcame US sanctions |url=https://www.technologyreview.com/2025/01/24/1110526/china-deepseek-top-ai-despite-sanctions/ |url-status=live |archive-url=https://web.archive.org/web/20250125180427/https://www.technologyreview.com/2025/01/24/1110526/china-deepseek-top-ai-despite-sanctions/ |archive-date=25 January 2025 |access-date=3 February 2025 |website=MIT Technology Review}}][{{cite journal |last1=Guo |first1=Daya |last2=Yang |first2=Dejian |last3=Zhang |first3=Haowei |last4=Song |first4=Junxiao |last5=Wang |first5=Peiyi |last6=Zhu |first6=Qihao |last7=Xu |first7=Runxin |last8=Zhang |first8=Ruoyu |last9=Ma |first9=Shirong |last10=Bi |first10=Xiao |last11=Zhang |first11=Xiaokang |last12=Yu |first12=Xingkai |last13=Wu |first13=Yu |last14=Wu |first14=Z. F. |last15=Gou |first15=Zhibin |date=18 September 2025 |title=DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning |journal=Nature |volume=645 |issue=8081 |pages=633–638 |bibcode=2025Natur.645..633G |doi=10.1038/s41586-025-09422-z |pmc=12443585 |pmid=40962978 |last16=Shao |first16=Zhihong |last17=Li |first17=Zhuoshu |last18=Gao |first18=Ziyi |last19=Liu |first19=Aixin |last20=Xue |first20=Bing |last21=Wang |first21=Bingxuan |last22=Wu |first22=Bochao |last23=Feng |first23=Bei |last24=Lu |first24=Chengda |last25=Zhao |first25=Chenggang |last26=Deng |first26=Chengqi |last27=Ruan |first27=Chong |last28=Dai |first28=Damai |last29=Chen |first29=Deli |last30=Ji |first30=Dongjie |last31=Li |first31=Erhang |last32=Lin |first32=Fangyun |last33=Dai |first33=Fucong |last34=Luo |first34=Fuli |last35=Hao |first35=Guangbo |last36=Chen |first36=Guanting |last37=Li |first37=Guowei |last38=Zhang |first38=H. |last39=Xu |first39=Hanwei |last40=Ding |first40=Honghui |last41=Gao |first41=Huazuo |last42=Qu |first42=Hui |last43=Li |first43=Hui |last44=Guo |first44=Jianzhong |last45=Li |first45=Jiashi |last46=Chen |first46=Jingchang |last47=Yuan |first47=Jingyang |last48=Tu |first48=Jinhao |last49=Qiu |first49=Junjie |last50=Li |first50=Junlong |last51=Cai |first51=J. L. |last52=Ni |first52=Jiaqi |last53=Liang |first53=Jian |last54=Chen |first54=Jin |last55=Dong |first55=Kai |last56=Hu |first56=Kai |last57=You |first57=Kaichao |last58=Gao |first58=Kaige |last59=Guan |first59=Kang |last60=Huang |first60=Kexin |last61=Yu |first61=Kuai |last62=Wang |first62=Lean |last63=Zhang |first63=Lecong |last64=Zhao |first64=Liang |last65=Wang |first65=Litong |last66=Zhang |first66=Liyue |last67=Xu |first67=Lei |last68=Xia |first68=Leyi |last69=Zhang |first69=Mingchuan |last70=Zhang |first70=Minghua |last71=Tang |first71=Minghui |last72=Zhou |first72=Mingxu |last73=Li |first73=Meng |last74=Wang |first74=Miaojun |last75=Li |first75=Mingming |last76=Tian |first76=Ning |last77=Huang |first77=Panpan |last78=Zhang |first78=Peng |last79=Wang |first79=Qiancheng |last80=Chen |first80=Qinyu |last81=Du |first81=Qiushi |last82=Ge |first82=Ruiqi |last83=Zhang |first83=Ruisong |last84=Pan |first84=Ruizhe |last85=Wang |first85=Runji |last86=Chen |first86=R. J. |last87=Jin |first87=R. L. |last88=Chen |first88=Ruyi |last89=Lu |first89=Shanghao |last90=Zhou |first90=Shangyan |last91=Chen |first91=Shanhuang |last92=Ye |first92=Shengfeng |last93=Wang |first93=Shiyu |last94=Yu |first94=Shuiping |last95=Zhou |first95=Shunfeng |last96=Pan |first96=Shuting |last97=Li |first97=S. S. |last98=Zhou |first98=Shuang |last99=Wu |first99=Shaoqing |display-authors=1 |last100=Yun |first100=Tao |last101=Pei |first101=Tian |last102=Sun |first102=Tianyu |last103=Wang |first103=T. |last104=Zeng |first104=Wangding |last105=Liu |first105=Wen |last106=Liang |first106=Wenfeng |last107=Gao |first107=Wenjun |last108=Yu |first108=Wenqin |last109=Zhang |first109=Wentao |last110=Xiao |first110=W. L. |last111=An |first111=Wei |last112=Liu |first112=Xiaodong |last113=Wang |first113=Xiaohan |last114=Chen |first114=Xiaokang |last115=Nie |first115=Xiaotao |last116=Cheng |first116=Xin |last117=Liu |first117=Xin |last118=Xie |first118=Xin |last119=Liu |first119=Xingchao |last120=Yang |first120=Xinyu |last121=Li |first121=Xinyuan |last122=Su |first122=Xuecheng |last123=Lin |first123=Xuheng |last124=Li |first124=X. Q. |last125=Jin |first125=Xiangyue |last126=Shen |first126=Xiaojin |last127=Chen |first127=Xiaosha |last128=Sun |first128=Xiaowen |last129=Wang |first129=Xiaoxiang |last130=Song |first130=Xinnan |last131=Zhou |first131=Xinyi |last132=Wang |first132=Xianzu |last133=Shan |first133=Xinxia |last134=Li |first134=Y. K. |last135=Wang |first135=Y. Q. |last136=Wei |first136=Y. X. |last137=Zhang |first137=Yang |last138=Xu |first138=Yanhong |last139=Li |first139=Yao |last140=Zhao |first140=Yao |last141=Sun |first141=Yaofeng |last142=Wang |first142=Yaohui |last143=Yu |first143=Yi |last144=Zhang |first144=Yichao |last145=Shi |first145=Yifan |last146=Xiong |first146=Yiliang |last147=He |first147=Ying |last148=Piao |first148=Yishi |last149=Wang |first149=Yisong |last150=Tan |first150=Yixuan |last151=Ma |first151=Yiyang |last152=Liu |first152=Yiyuan |last153=Guo |first153=Yongqiang |last154=Ou |first154=Yuan |last155=Wang |first155=Yuduan |last156=Gong |first156=Yue |last157=Zou |first157=Yuheng |last158=He |first158=Yujia |last159=Xiong |first159=Yunfan |last160=Luo |first160=Yuxiang |last161=You |first161=Yuxiang |last162=Liu |first162=Yuxuan |last163=Zhou |first163=Yuyang |last164=Zhu |first164=Y. X. |last165=Huang |first165=Yanping |last166=Li |first166=Yaohui |last167=Zheng |first167=Yi |last168=Zhu |first168=Yuchen |last169=Ma |first169=Yunxian |last170=Tang |first170=Ying |last171=Zha |first171=Yukun |last172=Yan |first172=Yuting |last173=Ren |first173=Z. Z. |last174=Ren |first174=Zehui |last175=Sha |first175=Zhangli |last176=Fu |first176=Zhe |last177=Xu |first177=Zhean |last178=Xie |first178=Zhenda |last179=Zhang |first179=Zhengyan |last180=Hao |first180=Zhewen |last181=Ma |first181=Zhicheng |last182=Yan |first182=Zhigang |last183=Wu |first183=Zhiyu |last184=Gu |first184=Zihui |last185=Zhu |first185=Zijia |last186=Liu |first186=Zijun |last187=Li |first187=Zilin |last188=Xie |first188=Ziwei |last189=Song |first189=Ziyang |last190=Pan |first190=Zizheng |last191=Huang |first191=Zhen |last192=Xu |first192=Zhipeng |last193=Zhang |first193=Zhongyu |last194=Zhang |first194=Zhen}}] This release made widely known how China had been embracing using and building more open AI systems as a way to reduce reliance on western software and gatekeeping as well as to help give its industries access to higher-powered AI more quickly.[{{Cite web |last=Bloom |first=Peter |date=12 February 2025 |title=DeepSeek: how China's embrace of open-source AI caused a geopolitical earthquake |url=http://theconversation.com/deepseek-how-chinas-embrace-of-open-source-ai-caused-a-geopolitical-earthquake-249563 |access-date=9 December 2025 |website=The Conversation |language=en-US}}] Projects based in China have since become more widely used around the world as well as they have closed at least some of the gap with leading proprietary American models.[{{Cite web |last=Huang |first=Raffaele |date=13 August 2025 |title=China's Lead in Open-Source AI Jolts Washington and Silicon Valley |url=https://www.wsj.com/tech/ai/chinas-lead-in-open-source-ai-jolts-washington-and-silicon-valley-ffdec83b |access-date=9 December 2025 |website=The Wall Street Journal |language=en-US}}][{{Cite web |last1=Cui |first1=Jasmine |last2=Perlo |first2=Jared |date=30 November 2025 |title=More of Silicon Valley is building on free Chinese AI |url=https://www.nbcnews.com/tech/innovation/silicon-valley-building-free-chinese-ai-rcna242430 |access-date=9 December 2025 |website=NBC News |language=en}}]
Since the release of OpenAI's proprietary ChatGPT model in late 2022, there have been only a few fully open (weights, data, code, etc.) large language models released. In September 2025, a Swiss consortium added to this short list by releasing a fully open model named [Apertus](/source/Apertus_(LLM)).[{{Cite web |last=Welle |first=Elissa |date=3 September 2025 |title=Switzerland releases an open-weight AI model |url=https://www.theverge.com/ai-artificial-intelligence/770646/switzerland-ai-model-llm-open-apertus |access-date=8 October 2025 |website=The Verge |language=en-US}}][{{Cite web |last=Allen |first=Matthew |date=2 September 2025 |title=Switzerland launches transparent ChatGPT alternative |url=https://www.swissinfo.ch/eng/swiss-ai/switzerland-launches-transparent-chatgpt-alternative/89929269 |access-date=8 October 2025 |website=[SWI swissinfo.ch](/source/SWI_swissinfo.ch) |language=en}}]
In December 2025, the [Linux Foundation](/source/Linux_Foundation) created the [Agentic AI Foundation](/source/Linux_Foundation), which assumed control of some open-source [agentic AI](/source/agentic_AI) protocols and other technologies created by OpenAI, [Anthropic](/source/Anthropic) and [Block](/source/Block%2C_Inc.).[{{Cite magazine |last=Knight |first=Will |title=OpenAI, Anthropic, and Block Are Teaming Up to Make AI Agents Play Nice |url=https://www.wired.com/story/openai-anthropic-and-block-are-teaming-up-on-ai-agent-standards/ |access-date=16 December 2025 |magazine=Wired |language=en-US |issn=1059-1028}}][{{Cite news |last=Claburn |first=Thomas |date=9 December 2025 |title=Linux Foundation aims to become the Switzerland of AI agents |url=https://www.theregister.com/2025/12/09/linux_foundation_agentic_ai_foundation/ |work=[The Register](/source/The_Register)}}]
Starting in November 2024, [Lightricks](/source/Lightricks) began releasing the [LTX](/source/LTX_(AI_Model)) video models as open weights.[{{Cite web |last=Lightricks |title=Lightricks Releases LTX-2: The First Complete Open-Source AI Video Foundation Model |url=https://www.prnewswire.com/news-releases/lightricks-releases-ltx-2-the-first-complete-open-source-ai-video-foundation-model-302593012.html |access-date=2026-04-23 |website=www.prnewswire.com |language=en}}][{{Cite web |last=Shulman |first=Sophie |date=2025-05-06 |title=Lightricks aims for its Deepseek moment with open-source AI video model |url=https://www.calcalistech.com/ctechnews/article/r1ckyydxle |access-date=2026-04-23 |website=ctech |language=en}}]
== Significance ==
The label ‘open-source’ can provide real benefits to companies looking to hire top talent or attract customers.[{{Cite magazine |last=Khalili |first=Joel |title=The Race to Build the DeepSeek of Europe Is On |url=https://www.wired.com/story/europe-race-us-deepseek-sovereign-ai/ |access-date=28 January 2026 |magazine=Wired |language=en-US |issn=1059-1028}}][{{Cite news |last=Desmarais |first=Anna |date=1 December 2025 |title=Europe is trying to write a new sovereign AI map. Here's how |url=https://www.euronews.com/next/2025/12/01/which-european-countries-are-building-their-own-sovereign-ai-to-compete-in-the-tech-race |access-date=29 January 2026 |website=Euronews |language=en}}]
== Licenses ==
As of 2025, a plurality (39%) of models released on [Hugging Face](/source/Hugging_Face) used the permissive [Apache 2.0](/source/Apache_2.0) License.[{{Cite journal |last1=Esteva |first1=Andre |last2=Robicquet |first2=Alexandre |last3=Ramsundar |first3=Bharath |last4=Kuleshov |first4=Volodymyr |last5=DePristo |first5=Mark |last6=Chou |first6=Katherine |last7=Cui |first7=Claire |last8=Corrado |first8=Greg |last9=Thrun |first9=Sebastian |last10=Dean |first10=Jeff |date=January 2019 |title=A guide to deep learning in healthcare |url=https://www.nature.com/articles/s41591-018-0316-z |journal=Nature Medicine |language=en |volume=25 |issue=1 |pages=24–29 |bibcode=2019NatMe..25...24E |doi=10.1038/s41591-018-0316-z |issn=1546-170X |pmid=30617335 |url-access=subscription}}] Open-source libraries have been used for medical imaging for tasks such as [tumor detection](/source/Cancer_screening), improving the speed and accuracy of diagnostic processes.[{{Cite book |last1=Ashraf |first1=Mudasir |title=International Conference on Innovative Computing and Communications |last2=Ahmad |first2=Syed Mudasir |last3=Ganai |first3=Nazir Ahmad |last4=Shah |first4=Riaz Ahmad |last5=Zaman |first5=Majid |last6=Khan |first6=Sameer Ahmad |last7=Shah |first7=Aftab Aalam |date=2021 |publisher=Springer |isbn=978-981-15-5113-0 |editor-last=Gupta |editor-first=Deepak |series=Advances in Intelligent Systems and Computing |volume=1165 |location=Singapore |pages=239–255 |language=en |chapter=Prediction of Cardiovascular Disease Through Cutting-Edge Deep Learning Technologies: An Empirical Study Based on TENSORFLOW, PYTORCH and KERAS |doi=10.1007/978-981-15-5113-0_18 |editor2-last=Khanna |editor2-first=Ashish |editor3-last=Bhattacharyya |editor3-first=Siddhartha |editor4-last=Hassanien |editor4-first=Aboul Ella |editor5-last=Anand |editor5-first=Sameer |editor6-last=Jaiswal |editor6-first=Ajay |chapter-url=https://link.springer.com/chapter/10.1007/978-981-15-5113-0_18}}][{{Cite journal |last1=Korshunova |first1=Maria |last2=Ginsburg |first2=Boris |last3=Tropsha |first3=Alexander |last4=Isayev |first4=Olexandr |date=25 January 2021 |title=OpenChem: A Deep Learning Toolkit for Computational Chemistry and Drug Design |url=https://pubs.acs.org/doi/10.1021/acs.jcim.0c00971 |journal=Journal of Chemical Information and Modeling |language=en |volume=61 |issue=1 |pages=7–13 |doi=10.1021/acs.jcim.0c00971 |issn=1549-9596 |pmid=33393291 |url-access=subscription}}]
=== Military ===
{{See also|Artificial intelligence arms race}}
Meta's Llama models, which have been described as open-source by Meta, were adopted by U.S. defense contractors like [Lockheed Martin](/source/Lockheed_Martin) and [Oracle](/source/Oracle_Corporation) after unauthorized adaptations by Chinese researchers affiliated with the [People's Liberation Army](/source/People's_Liberation_Army) (PLA) came to light.[{{Cite news |last1=Pomfret |first1=James |last2=Pang |first2=Jessie |last3=Pomfret |first3=James |last4=Pang |first4=Jessie |date=1 November 2024 |title=Exclusive: Chinese researchers develop AI model for military use on back of Meta's Llama |url=https://www.reuters.com/technology/artificial-intelligence/chinese-researchers-develop-ai-model-military-use-back-metas-llama-2024-11-01/ |access-date=16 November 2024 |work=Reuters |language=en}}][{{Cite web |last=Roth |first=Emma |date=4 November 2024 |title=Meta AI is ready for war |url=https://www.theverge.com/2024/11/4/24287951/meta-ai-llama-war-us-government-national-security |access-date=16 November 2024 |website=The Verge |language=en}}] The [Open Source Initiative](/source/Open_Source_Initiative) and others have contested Meta's use of the term ''open-source'' to describe Llama, due to Llama's license containing an [acceptable use policy](/source/acceptable_use_policy) that prohibits use cases including non-U.S. military use.[{{Cite web |last=Piper |first=Kelsey |date=2 February 2024 |title=Should we make our most powerful AI models open source to all? |url=https://www.vox.com/future-perfect/2024/2/2/24058484/open-source-artificial-intelligence-ai-risk-meta-llama-2-chatgpt-openai-deepfake |access-date=16 December 2025 |website=Vox |language=en-US}}] Its authors propose that health-care institutions, academic researchers, clinicians, patients and technology companies worldwide should collaborate to build open-source models for health care of which the underlying code and base models are easily accessible and can be fine-tuned freely with own data sets.[{{Cite journal |last=Dean |first=Jeffrey |date=1 May 2022 |title=A Golden Decade of Deep Learning: Computing Systems & Applications |url=https://direct.mit.edu/daed/article/151/2/58/110623/A-Golden-Decade-of-Deep-Learning-Computing-Systems |journal=Daedalus |language=en |volume=151 |issue=2 |pages=58–74 |doi=10.1162/daed_a_01900 |issn=0011-5266 |doi-access=free}}] The open-source nature of these platforms also facilitates rapid iteration and improvement, as contributors from across the globe can propose modifications and enhancements to existing tools.[{{Cite journal |last1=Hassri |first1=Myftahuddin Hazmi |last2=Man |first2=Mustafa |date=7 December 2023 |title=The Impact of Open-Source Software on Artificial Intelligence |url=https://journal.umt.edu.my/index.php/jmsi/article/view/471 |journal=Journal of Mathematical Sciences and Informatics |language=en |volume=3 |issue=2 |doi=10.46754/jmsi.2023.12.006 |issn=2948-3697 |url-access=subscription |doi-access=free}}][{{cite magazine |last1=Solaiman |first1=Irene |author-link=Irene Solaiman |date=24 May 2023 |title=Generative AI Systems Aren't Just Open or Closed Source |url=https://www.wired.com/story/generative-ai-systems-arent-just-open-or-closed-source/ |url-status=live |archive-url=https://web.archive.org/web/20231127053255/https://www.wired.com/story/generative-ai-systems-arent-just-open-or-closed-source/ |archive-date=27 November 2023 |access-date=20 July 2023 |magazine=Wired}}] This can help to create an ecosystem for other businesses to sell services on top of.[{{Cite news |last=Lin |first=Belle |date=21 March 2024 |title=Open-Source Companies Are Sharing Their AI Free. Can They Crack OpenAI's Dominance? |url=https://www.wsj.com/articles/open-source-companies-are-sharing-their-ai-free-can-they-crack-openais-dominance-26149e9c |access-date=16 December 2025 |work=Wall Street Journal |language=en-US |issn=0099-9660}}]
=== Transparency ===
thumb|A video about the importance of transparency of AI in medicine
One benefit of open-source AI is the increased transparency it offers compared to closed-source alternatives.[{{Cite web |last=Gujar |first=Praveen |title=Council Post: Building Trust In AI: Overcoming Bias, Privacy And Transparency Challenges |url=https://www.forbes.com/councils/forbestechcouncil/2024/11/19/building-trust-in-ai-overcoming-bias-privacy-and-transparency-challenges/ |access-date=27 November 2024 |website=Forbes |language=en}}]{{Better source needed|reason=The current source is insufficiently reliable ([WP:NOTRS](/source/WP%3ANOTRS)).|date=December 2025}}
== Concerns ==
=== Quality and security ===
Open sourced models have fewer ways to prevent them from being used for malicious activities.[{{Cite web |last=Sandbrink |first=Jonas |date=7 August 2023 |title=ChatGPT could make bioterrorism horrifyingly easy |url=https://www.vox.com/future-perfect/23820331/chatgpt-bioterrorism-bioweapons-artificial-inteligence-openai-terrorism |access-date=14 November 2024 |website=Vox |language=en-US}}][{{Cite web |last=O'Brien |first=Matt |date=30 July 2024 |title=White House says no need to restrict open-source AI, for now |url=https://www.pbs.org/newshour/nation/white-house-says-no-need-to-restrict-open-source-ai-for-now |access-date=14 November 2024 |website=Associated Press |publisher=PBS News |language=en-us}}] though a number of experts in 2024 seemed more concerned about future advances than present-day capabilities.[{{Cite web |last=Kahn |first=Jeremy |title=Why China's open source AI models are eating the world |url=https://fortune.com/2025/11/25/outside-the-u-s-and-europe-the-momentum-of-chinas-open-source-ai-models-is-plain-to-see/ |access-date=29 January 2026 |website=Fortune |language=en}}]
=== Training costs ===
The cost of training datasets for fully open-sourced models can be prohibitively expensive for many projects.