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.
Naver Cloud, NC AI cut from national AI foundation model race
Second Vice Science Minister Ryu Je-myung speaks during a press briefing to announce the first evaluation results of the national artificial intelligence (AI) foundation model project at the government complex in Seoul, Thursday. Yonhap
By Lee Gyu-lee
LG AI Research, SK Telecom, Upstage advances with one new recruit to join second round
Two consortia, led by Naver Cloud and NC AI, have been eliminated from the government-led national artificial intelligence (AI) foundation model project after the first round of evaluation.
The Ministry of Science and ICT announced the results of the first evaluation on Thursday, revealing that only three teams — LG AI Research, SK Telecom and Upstage — advanced to the second round.
The ministry said that the evaluation for the first stage involved three assessments: benchmark performance, expert evaluation and user experience.
LG AI Research’s team received the highest overall score with 90.2 points, while the average score of the five shortlisted teams was 79.7 points. LG's model dominated in all three assessments, scoring 33.6 out of 40 points in benchmark, 31.6 out of 35 points in expert review and a perfect 25 points in user experience.
The ministry explained that Naver Cloud was disqualified for failing to meet the project’s requirement for technology originality.
Naver said it accepts the ministry's decision and will not seek to rejoin the project, despite the option to appeal or reapply in the second round.
After the five consortia unveiled their models last month ahead of the evaluation, some teams faced plagiarism controversies over the use of open-source code and encoders from other models.
The ministry explained that the technology's originality was assessed for its technical, policy and ethical perspectives. Technically, an independent model must be trained end-to-end from scratch with self-developed architectures, data and learning processes, even when open-source tools are used.
From a policy and ethical perspective, the model must guarantee sovereign control without any external licensing constraints, while also ensuring transparency and compliance with open-source licensing rules.
The ministry explained that although Naver Cloud used licensing-free open-source video and audio encoders, the company failed to meet the minimum requirement to train and develop the model with newly initialized weights. Weights are the learned settings that determine how AI processes information and makes decisions.
“Even if an existing open model was used, there should have been concrete, demonstrable efforts to completely reset the weights and rebuild them using data secured independently by the team, with such experience proven and verified,” Second Vice Science Minister Ryu Je-myung said during a press briefing at the government complex in Seoul, Thursday.
People visit LG AI Research's booth during a presentation event for the national artificial intelligence foundation model project at Coex in southern Seoul, Dec. 30, 2025. Yonhap
The ministry said the two eliminated teams will have 10 days to file an appeal and announced plans to recruit an additional team for the second round, as only three teams advanced — short of the four originally planned.
The ministry added that, in addition to Naver Cloud and NC AI, all consortia that initially applied for the project, as well as other qualified companies, are eligible to compete for the open spot.
The ministry explained the evaluation process, detailing each of the three assessments. The evaluation involved National Information Society Agency (NIA) benchmarks covering math, reasoning, long-text understanding, reliability and safety, as well as 13 widely recognized tests, comparing each model against state-of-the-art global standards.
SK Telecom and LG AI Research tied for the top score in the NIA benchmarks with 9.2 out of 10 points, while LG led the global common benchmarks with 14.4 out of 20 points. Upstage and LG both achieved perfect scores in the target-specific benchmarks.
“We are honored to have advanced to the second phase of the national AI foundation model project,” an SK Telecom official said.
“In the second phase, we plan to add multimodal capabilities and carry out additional training for our model. Over the longer term, we are aiming to scale the model to the trillion-parameter level.”
Expert evaluation was conducted by a panel of 10 external AI specialists from the industry, academia and research institutions, carrying out an in-depth review of the teams’ technical reports and AI model training logs to examine development processes, technical capabilities and originality.
The average score of the five teams was 28.5 out of 35, while the LG AI Research team notched 31.6.
For the user experience assessment, a group of 49 AI-specialized users, including AI startup founders, evaluated AI service websites built using the teams’ models for real life usability and cost efficiency.
LG AI Research scored a perfect 25 points, while the average was 20.76.