From tech giants to startups, South Korean players develop large language models tailored to their own language and culture, ready to compete with global heavy weights such as Openai and Google.
Last month, the nation launched its most ambitious superb AI initiative to date and promised 530 billion (about $ 390 million) to five local businesses building large basic models.
The move emphasizes Seoul’s desire to cut dependence on foreign AI technologies in the hope of strengthening national security and keeping a tighter control over data in the AI ​​era.
The organizations chosen by the Minister of Science, and I competed were LG AI Research, SK Telecom, Naver Cloud, NC AI and Start -up -Upstage.
Every sixth month, the government will review the first cohort’s progress, cut underprestors and continue to finance the front trunners until there are only two left to lead the country’s sovereign AI drive.
Each player brings another benefit to South Korea’s AI Race. Techcrunch talked to several of the chosen companies about how they plan to take on Openai, Google, Anthropic and the rest on their home ground. NC AI refused to comment.
LG AI Research: Exaone
LG AI Research, F&U Unit for the South Korean Giant LG Group, offers Exaone 4.0, a hybrid reasoning AI model. The latest version blends broad language processing with the advanced reasoning features that were first introduced in the company’s previous Exaone Deep Model.
TechCrunch -event
San Francisco
|
27-29. October 2025
Exaone 4.0 (32B) already scores reasonably good moditors on Artificial Analysis’s Intelligence Index Benchmark (like Upstage’s Solar Pro2). But it plans to improve and move up the rows through its deep access to the real world’s industrial data, rooted from biotech to advanced materials and manufacture.
It connects this data focusing on refining the data before being fed to the models to train. Instead of chasing clean scale, LG wants to make the whole process more intelligent so that its AI can deliver real, practical value that goes beyond what general models can offer. “This is our basic approach,” Co-Head Honglak Lee told TechCrunch.
LG improves its models through well -known tactics: Offers them through APIs and then uses the real world data generated by users of these services to train the model to improve.
“As LG’s models are improved, our partners can provide better services, which in turn generate greater economic value and even richer data,” he said.
However, the Intad of Chasing focuses massive GPUS clusters, LG AI research on efficiency, gets the most out of each chip and creates industry-specific models, he said. The goal is not to leave the global giants, but to outlast them with high -performance, yet more effective AI.
South Korea’s Telco giant SK Telecom (Skt) launched its personal AIA agent A. (pronounced A-DOT) far back in late 2023 and just rolled its big big language model, AX, in July.
Built on top of the Chinese Open Source model from Alibaba Cloud, Qwen 2.5, AX 4.0 comes in two models, a hefty 72 billion parameter version and a lighter 7B version.
SK says that AX 4.0 Korean Input approx. 33% more efficiently than the GPT-4o did, which emphasized its language edge. (Openais GPT 5.0 comparison data is not available.) SKT also opens its AX 3.1 models earlier this summer. Meanwhile, A. Service features such as AI calls and auto-generated notes. From August 2025, about 10 million subscribers have already been drawn.
SK’s Edge is its versatility because it has access to information from its telecommunications lover that ran from navigation to taxi-hailing.
“SK Telecom’s role is to act as a bridge between advanced model study and effect in the real world. With our telecom infrastructure, comprehensive user base and proven service such as A., we bring AI directly into everyday life, whether in customer service or manufacture,” Kim, head of Foundation Model Office at SK Telecom, Techcrunch said.
SK Telecom also invests in AI-Infrastructure, GPUAAAS, South Korea’s large GPU-based service and builds a new hyperscale AI data center with AWS. Whatever it is missing, it collaborates to get.
“We are building a full-staple ecosystem with Korean AI chip maker uprising, securing trusted data partnerships through work with the government and universities and promoting a global research network,” Kim said. “It includes projects such as our collaboration with MIT (MGAIC), which apply to foundation models for advanced production and battery and semiconductor innovation.”
Naver Sky: Hyperclova x
Naver Cloud, Cloud Services arm from South Korea’s leading internet company, introduced its large language model, Hyperclova, in 2021. Two years later, it revealed an upgraded version, Hyperclova X, along with new products driven by the technology: Clova X, generative AIT-driven search engine located as a rival to Microsoft’s Copilot Google’s AI-Overview. It also revealed this year its multimodal reasoning AI model, Hyperclove X Think.
Naver Cloud believes that the true power of the LLMs is to act as “connectors” connecting older systems and silent services to improve the utility, according to a naver spokesman.
Naver stands out as Korea’s only company – and one of the few in the world – who can really claim to have an “ai full stack.” It builds its Hyperclova X model from scratch and runs the massive data centers, cloud services, AI platforms, applications and consumer services that bring the technology to life, the spokesman explained.
Like Google – but set to South Korea – Naver is embedding its AI in core services such as search, shopping, maps and finance. Its advantage is data in the real world. For example, it’s AI Shopping Guide, offers recommendations based on what People News Worm. Other services included Clova Studio, letting companies build up custom generative AI, and Clova CareCall, an A-Pown-Check-in service intended for seniors living alone.
The Naver spokesman says that Besting Global AI giants like Openai and Google are related to two things: to perfect his “recipe” to models and secure the capital to scale them. Yet, rather than chasing size, the company emphasizes sophistication and argues that its AI is already globally complete in comparable scales.
Upstage’s Solar Pro 2
Upstage is the only start -up competing in the project. Its Solar Pro 2 model launched last July was the first Korean model to be recognized as a border model by artificial analysis that put it in the ring with Openai, Google, Meta and Anthropic, according to Soon Il Kwon, Executive Vice President.
While most Frontier models have 100 billion to 200 billion parameters, the Solar Pro 2-with only 31 billion performs better for South Koreans and is more cost-effective, Kwon told Techcrunch.
“Solar Pro 2 has surpassed global models on larger Korean benchmarks. With this project, Upstage aims to achieve a Korean language benefit of 105% of the global standard,” Kwon said.
Upstage aims to differentiate themselves by focusing on real business impact, not just benchmarks, he said. So it is to develop specialized models for industries such as Finance, Law and Medicine, while it is pushing to build a Korean AI-ecosystem led by “AI-native” startups.