Lee Gyu-lee is a business writer at The Korea Times, focusing primarily on IT & telecommunications, the Ministry of Trade, Industry and Energy and KOTRA. Prior to this, she has covered a wide range of cultural news, from film, television and K-pop to lifestyle and fashion.
Consortia unveil models for national AI project
Minister of Science and ICT Bae Kyung-hoon speaks during a presentation event for the national AI foundation model project at Coex in southern Seoul, Tuesday. Courtesy of Ministry of Science and ICT
By Lee Gyu-lee
Five consortia participating in Korea’s national artificial intelligence (AI) foundation model project unveiled their models as the first-stage outcome, marking a major step forward in the nation’s ambition to become one of the world’s leading AI powerhouses.
During a presentation event for the project at Coex in southern Seoul Tuesday, five teams — led by Naver Cloud, Upstage, SK Telecom, LG AI Research and NC AI — introduced their initial models and road maps for further development.
The project, which kicked off in August, aims at building independent, globally competitive foundation AI models, with the government investing around 200 billion won ($136 million) to support the consortia with high-performance graphics processing units (GPUs), datasets and R&D personnel.
The government will conduct evaluations twice a year, removing one team each time until only two finalists remain by the first half of 2027.
Drawing more than 1,000 industry leaders and researchers, the event featured team presentations, along with booths to demonstrate developments to the general public.
“Many countries around the world, including the United States and China, are showing great interest in this project, where the government and private sector join forces to develop AI foundation models. They will also be paying close attention to the outcomes,” Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon said.
“Building on these results, we will go beyond developing an AI model to create great services, platforms and globally scalable products in collaboration with the participating companies.”
People visit LG AI Research's booth during a presentation event for the national AI foundation model project at Coex in southern Seoul, Tuesday. Yonhap
Ha Jung-woo, senior presidential secretary for AI future planning, noted the importance of the project in building the foundations of Korea’s AI ecosystem.
“This project is fundamentally about building the basic strength that underpins our competitiveness in AI algorithms and AI technologies,” he said. "The government will actively support not only AI and energy infrastructure, but also our fundamental technological capabilities and operational expertise so they can be fully applied across industries, the public sector and society.”
Naver Cloud showcased HyperCLOVA X OMNI, a Korean-optimized multimodal model designed to understand and process text, images and audio within a unified framework.
The model is optimized for deep reasoning and applies holistic learning across diverse tasks so that one model can handle everything from everyday conversational agents to complex enterprise workflows without switching between separate models.
Upstage introduced Solar Open, designed as an agentic reasoning-focused model with strong multilingual capabilities and support for industry-specific applications in finance, law, health care and e-commerce.
SK Telecom revealed A.X K1, a massive 500-billion-scale parameter AI model. Parameters are learned factors a model relies on to interpret data and are widely regarded as an indicator of the model’s scale or capacity.
The model combines large-scale computational power with advanced reasoning abilities to handle complex real-world tasks across industries.
Participants try SK Telecom-led consortium's artificial intelligence (AI) model A.X K1 at a presentation event for the national AI foundation model project at Coex in Southern Seoul, Tuesday. Courtesy of SK Telecom
LG AI Research presented K-EXAONE, its next-generation frontier model built with 236 billion parameters to achieve both efficiency and scalability. By integrating a unique mixture-of-experts structure and hybrid attention technology, K-EXAONE significantly reduces computation and memory requirements while improving reasoning speed.
The model outperformed major open-weight rivals such as OpenAI’s gpt-oss-120b and Alibaba’s Qwen3-235B, achieving 104 percent of their benchmark performance.
NC AI focused on VAETKI, a multimodal, industry-specialized foundation model tailored to apply across diverse sectors from manufacturing and defense to retail and entertainment. The model aims to deliver cost-efficient and secure AI optimized for on-premise and domain-specific environments.
The ministry plans to conduct an evaluation in January, reviewing both technical achievements and future road maps for the first elimination.