South Korea has established itself rapidly as an innovator of large language models, driven by government strategic investments, corporate R&D, and collaborations with open-source software to develop models specifically tailored for Korean Language Processing and domestic applications. This helps reduce dependence on foreign AI technology, improves data privacy and supports sectors such as healthcare, education and telecommunications.
Government-Backed Push for Sovereign AI
In 2025, the Ministry of Science and ICT initiated a 240 billion won program, selecting five consortia—led by Naver Cloud, SK Telecom, Upstage, LG AI Research, and NC AI—to develop sovereign LLMs capable of operating on local infrastructure.
The Ministry of Food and Drug Safety guidelines on text-generating artificial intelligence in medicine, which will mark the world’s first framework of this kind by the early years of 2025.
Innovating in the Corporate Sector and at Academic Institutions
SK Telecom launched AX 3.1 Lite – a 7-billion parameter model that was trained completely from scratch with 1.65 trillion multilingual symbols, with a Korean focus. This model has a performance level of approximately 96% for KMMLU2 in Korean language reasoning, and 102% for CLIcK3 cultural understanding compared with larger models.
In June 2025 Naver will release HyperClova X Think, a new version of its HyperClova Series that enhances Korean search and conversations.
Upstage’s Solar Pro 2 has been ranked as the top Korean product on Frontier LM Intelligence, displaying a performance that is comparable to much larger international products.
LG AI Research released Exaone 4.0, a design with 30 billion parameters that performs well in benchmarks around the world.
Seoul National University Hospital created Korea’s first Medical LLM. It was trained using 38 million clinical records that were de-identified. The hospital scored 86.2% in the Korean Medical Licensing Examination, compared with the average human score of 79.7%.
Mathpresso and Upstage have collaborated to develop MATH GPT. It is a small LLM with 13 billion parameters that exceeds GPT-4 on mathematical benchmarks, by using 488 percent accuracy instead of the previous GPT-4, but requires significantly less computing resources.
By continuously pretraining Korean datasets, open-source initiatives such as Polyglot-Ko and Gecko-7B can address the gaps in linguistics.
Tech Trends
Korean developers are focusing on efficiency. They optimize token to parameter ratios using Chinchilla-scaled models, allowing 7 to 30 trillion-parameter models compete with their larger Western counterparts even though resources are limited.
As seen with the MATH GPT and medical LLM at Seoul National University Hospital, domain-specific adaptations produce superior results for targeted areas.
Benchmarks such as KMMLU2, the CLIcK3 leaderboard for cultural relevancy, and Frontier LM confirm that we are on par with other advanced global systems.
Market Outlook
South Korean LLM is predicted to increase from 182.4 millions USD in 2020 to 1 278.3 by 2030. The growth will be driven by virtual assistants (chatbots), sentiment analysis, and chatbots. The integration of edge computing LLMs into telecom companies supports reduced latency, enhanced data security and initiatives such as the AI Infrastructure Superhighway.
South Korean Large Language Models Mentioned
| # | Model | Developer / Lead Institution | Parameter count | The Notable Focus |
|---|---|---|---|---|
| 1 | AX 3.1 Light | SK Telecom | 7 Billion | Korean mobile and device processing |
| 2 | AX 4.0 Lite | SK Telecom | 72 Billion | Scalable sovereign applications |
| 3 | HyperClova X Think | Naver | Estimated 204 Billion | Korean language search and dialog |
| 4 | Solar Pro 2 | The Upstage | The estimated value of the 30 Billion (est. | Leaderboards for global efficiency |
| 5 | MATH GPT | Mathpresso + Upstage | 13 Billion | Mathematics specialization |
| 6 | Exaone 4 | LG AI Research | The 30 Billion Dollars | Multimodal AI capabilities |
| 7 | Polyglot-Ko | EleutherAI + KIFAI | From 1.3 billion to 12.80 billion | Korean only open source training |
| 8 | Gecko-7B | Beomi community | 7 Billion | Korean language training should be continuous |
| 9 | SNUH Medical LLM | Seoul National University Hospital | undisclosed (~15B est.) | Support for clinical and medical decisions |
The South Korean approach in creating AI models with cultural relevance and efficiency is highlighted by these developments. They strengthen South Korea’s position on the global tech landscape.
Sources:
- https://www.cnbc.com/2025/08/08/south-korea-to-launch-national-ai-model-in-race-with-us-and-china.html
- https://www.forbes.com/sites/ronschmelzer/2025/07/16/sk-telecom-releases-a-korean-sovereign-llm-built-from-scratch/
- https://www.kjronline.org/pdf/10.3348/kjr.2025.0257
- https://www.rcrwireless.com/20250714/ai/sk-telecom-ai-3
- https://huggingface.co/skt/A.X-3.1-Light
- https://www.koreaherald.com/article/10554340
- http://www.mobihealthnews.com/news/asia/seoul-national-university-hospital-builds-korean-medical-llm
- https://www.chosun.com/english/industry-en/2024/05/03/67DRPIFMXND4NEYXNFJYA7QZRA/
- https://huggingface.co/blog/amphora/navigating-ko-llm-research-1
- https://www.grandviewresearch.com/horizon/outlook/large-language-model-market/south-korea

