LG AI Research has had three of its models—Exaone Deep, Exaone Path 2.0, and Exaone 4.0 listed as “Notable AI Models” by U.S. (Image courtesy of LG AI Research)
SEOUL, July 24 (Korea Bizwire) — South Korea’s domestic AI industry is demonstrating impressive progress, producing globally recognized models despite limited access to computing infrastructure and top-tier talent. Industry analysts say the recent string of successes strengthens Seoul’s broader goal of becoming a top-three global AI power by developing its own sovereign AI capabilities.
According to the AI sector on Wednesday, LG AI Research has had three of its models—Exaone Deep, Exaone Path 2.0, and Exaone 4.0—listed as “Notable AI Models” by U.S.-based nonprofit Epoch AI, a designation often cited by Stanford University’s Human-Centered AI Institute.
LG now has four entries on the list, including Exaone 3.5, released last year. The list includes 965 global models and research papers, with 16 South Korean models featured, including Naver’s HyperCLOVA 204B.
Another standout is Upstage, which recently unveiled Solar Pro 2, an AI model designed for complex problem-solving through logical reasoning. The model ranked 12th globally in Artificial Analysis’s intelligence index, outperforming other sovereign models like China’s DeepSeek V3 and France’s Mistral Small. Upstage also ranked 8th globally among companies, placing it alongside major AI developers like OpenAI, Google, and Meta.
The achievement drew attention on social media when Upstage CEO Kim Sung-hoon shared the model’s ranking on X (formerly Twitter), prompting a reply from Elon Musk, founder of Tesla and xAI. Musk noted that xAI’s Grok remains at the top and is “improving rapidly,” to which Kim responded, “Congrats on staying at No.1 — but it won’t last long.”
Observers noted the symbolic weight of xAI—backed by over $2.6 billion in infrastructure investment—responding to a more compact but efficient Korean model built with far fewer resources.
(Image from Upstage CEO Kim Sung-hoon X Twitter)
Naver Cloud, developer of HyperCLOVA X, also reported strong results. The company said its Seed 3B model performed better than similarly sized competitors like Qwen 2.5-VL-3B (China) and Gemma3-4B (Google) in nine video-related benchmarks, including both Korean and English language tasks.
The model is also gaining traction on open-source platform Hugging Face, with over 1.3 million downloads as of July 21, placing it in the top 0.02% of all models.
“High benchmark scores are important, but real-world adoption hinges on cost-effective deployment,” a Naver official noted, pointing to the model’s growing presence in industry use.
A government official added that the performance of Korea’s lightweight, inference-specialized AI models reflects meaningful progress, supported by a growing open-source ecosystem.
With strategic investment and continued development, South Korea appears increasingly well-positioned to stake its claim in the competitive global AI race.
Kevin Lee (kevinlee@koreabizwire.com)
