{"id":556,"date":"2026-06-26T16:21:26","date_gmt":"2026-06-26T07:21:26","guid":{"rendered":"https:\/\/blog.hancom.com\/percent-ec-percent-98-percent-a4-percent-ed-percent-94-percent-88-percent-ec-percent-86-percent-8c-percent-ec-percent-8a-percent-a4-ai-percent-eb-percent-9e-percent-80-percent-eb-percent-ac-percent-b4\/"},"modified":"2026-07-13T15:13:43","modified_gmt":"2026-07-13T06:13:43","slug":"what-is-open-source-ai","status":"publish","type":"post","link":"https:\/\/blog.hancom.com\/en\/what-is-open-source-ai\/","title":{"rendered":"What Is Open Source AI? From Concepts to Domestic and International Corporate Cases"},"content":{"rendered":"\n
\ud83e\uddd1\u200d\ud83d\udcbb Summary:This article first explores the concept of open source\u2014software whose source code is publicly available for anyone to freely use, modify, and redistribute\u2014and the representative case of Hancom's 'OpenDataLoader PDF v2.0,' which met these criteria and reached the #1 spot on GitHub's overall trending list. Next, we examine the four freedoms and three disclosure requirements necessary to be recognized as Open Source AI, and introduce DeepSeek R1, which is considered one of the cases most faithful to these standards. Finally, we look at the trend toward open AI adoption among domestic and international companies such as OpenAI, Meta, Google, and LG.  <\/pre>\n\n
<\/div>\n\n
“We were on the wrong side of history.”<\/p>\n\n
These are the words OpenAI CEO Sam Altman posted on the online community Reddit. The ‘wrong side’ he referred to is OpenAI’s ‘Closed AI’ policy. OpenAI has long maintained a closed model approach, refusing to disclose ChatGPT’s source code, citing security and model safety concerns.   <\/p>\n\n
This statement represents a public acknowledgment by the leader of a company that has pioneered the AI industry that there were issues with the company’s strategic direction. Indeed, the AI market has recently seen a rapid shift toward open AI. So what exactly is open source, the core condition of open AI, and how are companies participating in this trend? <\/strong><\/p>\n\n
This article explores the concept of open source, the criteria for open source AI, representative cases, and the open AI strategies of global big tech companies and domestic enterprises.<\/p>\n\n
<\/div>\n\n
What Is Open Source? <\/strong><\/h2>\n\n
Open source refers to software whose source code is publicly available for anyone to freely use, modify, and redistribute, as well as a development methodology built on open collaboration.<\/pre>\n\n
<\/div>\n
\n
\"Open<\/figure>\n<\/div>\n
<\/div>\n\n
The Open Source Initiative (OSI)<\/a>, which governs the global standard for open source principles, specifies that software must meet ten criteria\u2014including free redistribution, source code disclosure, and permission for derivative works\u2014to be recognized as open source. This openness has enabled productive exchanges of ideas, allowing developers worldwide to collectively review and improve publicly available source code and create new software based on it.  <\/p>\n\n
<\/div>\n\n
Representative Open Source Case: Hancom ‘OpenDataLoader’ PDF v2.0 <\/h3>\n\n
A representative case that actually meets open source criteria is Hancom’s ‘OpenDataLoader’ PDF v2.0. <\/p>\n\n
PDF is the most widely used document format for AI training worldwide. However, its complex internal structure makes data extraction challenging, and it has been identified as a major bottleneck in AI development. OpenDataLoader PDF v2.0 is a technology designed to solve this problem by decomposing complex PDF documents into text, tables, and images, converting them into a format that AI can process directly.  <\/p>\n\n
OpenDataLoader PDF v2.0 applies the OSI-officially certified Apache 2.0 License<\/a>, opening it up for commercial use, and realizes the core open source value of transparency by publicly releasing not only the source code but also benchmark test data and reproduction code via GitHub. As a result, it achieved GitHub’s overall trending #1 ranking<\/a> within one week of release and ranked first in all categories of global PDF extraction performance benchmarks, receiving substantial validation from the global developer community.  <\/p>\n\n
<\/div>\n\n
Why Open Source Disclosure Is Difficult in AI<\/strong><\/h2>\n\n
Open source disclosure is difficult in AI because the components that must be disclosed are far more complex<\/strong> than in software. <\/p>\n\n
While software can be recognized as open source simply by disclosing the source code, AI requires disclosure of not only the source code but also the datasets used to train the AI, the various training models needed to create the final model, parameters including weights, and the computing environment capable of running the computational system\u2014far more components overall. <\/p>\n\n
<\/div>\n\n
What Is Open Source AI?<\/strong><\/h2>\n\n
Open source AI refers to an AI system that anyone can freely use for any purpose, examine its internal structure, modify, or redistribute.<\/pre>\n\n
<\/div>\n
\n
\"Open<\/figure>\n<\/div>\n
<\/div>\n\n
Open Source Initiative (OSI) Open Source AI Definition 1.0<\/h3>\n\n
As discussed earlier, defining open source in AI is not straightforward. To address this issue, the Open Source Initiative (OSI) released the first official definition of open source AI in October 2024.  <\/p>\n\n
<\/div>\n\n
Open Source AI Criteria<\/h4>\n\n
According to the Open Source AI Definition 1.0 (OSAID)<\/a> presented by OSI, recognition as open source AI requires guaranteeing four freedoms and meeting three disclosure requirements.<\/p>\n\n
What Must Be Guaranteed (Four Freedoms) <\/p>\n\n