Among the five elite teams participating in the government's "national AI model project," Naver Cloud and NC AI teams were eliminated in the first round on the 15th. The NCAI elite team was eliminated in the first-stage evaluation, and the Naver Cloud elite team was dropped on the grounds of originality. The national AI model project is a core government program that concentrates support for resources such as graphics processing units (GPU) and data, with the goal of building a uniquely Korean AI model to rank among the world's top three AI countries.
Second Vice Minister Ryu Jae-myung of the Ministry of Science and ICT stated accordingly at a briefing on the "independent AI foundation model project first-stage evaluation results" held at Government Complex Seoul on the afternoon of the 15th. As a result, the elite teams from Upstage, SK Telecom, and LG AI Research advanced to the second stage.
The first-stage evaluation consisted of △benchmark (40 points) △expert evaluation (35 points) △user evaluation (35 points). In the three categories, the elite teams from LG AI Research, Naver Cloud, SK Telecom, and Upstage, excluding NC AI, met the requirements. However, the Naver Cloud elite team was eliminated on the originality front.
The Ministry of Science and ICT said, "After evaluating the criteria for the independent AI foundation model project from technical, policy, and ethical perspectives, we determined that the Naver Cloud elite team's AI model does not meet the originality criteria," adding, "Expert evaluators also raised issues with originality limitations, leading to the decision to eliminate the Naver Cloud elite team."
The Ministry of Science and ICT added that it plans to select one additional elite team, as two teams, not one, were eliminated in the first round. Eligible applicants for the additional slot include consortia that applied to the initial project call, consortia not included among elite teams after the first-stage evaluation (the Naver Cloud and NC AI consortia), and other corporations. The ministry said, "The one additional elite team selected will be given opportunities to take on the development of the independent AI foundation model project, including support for graphics processing units (GPU) and data, and the designation 'K-AI company,' and we will swiftly proceed with the additional call for one elite team through administrative procedures."
◇ LG AI Research tops benchmark, expert, and user evaluations... NC AI is eliminated
In the first-stage evaluation, LG AI Research earned the highest scores across all areas.
First, for the "benchmark" score, the government conducted evaluations by the National Information Society Agency (NIA) △benchmark evaluation (10 points) △global common benchmark evaluation (20 points) △global individual benchmark evaluation (10 points). In the NIA benchmark evaluation, in addition to math, knowledge, and long-form comprehension, reliability and safety were also considered in collaboration with the AI Safety Research Institute. For the global common benchmark evaluation, 13 globally recognized benchmarks (spanning multiple areas such as agents, math, knowledge/reasoning, and instruction following) were selected and assessed. For the global individual benchmark evaluation, five benchmarks comparable to each elite team's global target model (SOTA level) were evaluated.
In the NIA benchmark evaluation, SK Telecom and LG AI Research each scored a top mark of 9.2 out of 10. In the global common benchmark evaluation, LG AI Research scored a top mark of 14.4 out of 20. In the global individual benchmark evaluation, Upstage and LG AI Research each scored a top mark of 10.0 out of 10. Combining these, LG AI Research earned the top overall benchmark score of 33.6. The five-team average was 30.4.
For the expert evaluation, a panel of 10 external AI experts from industry, academia, and research evaluated the materials submitted by each team and conducted an in-depth review over an extended period. They conducted a comprehensive assessment of △development strategy and technology △development outcomes and plans △ripple effects and contribution plans.
The evaluation panel analyzed each team's publicly released technical report, AI model training state log files, and more to assess the elite teams' technology development processes and capabilities (including originality). LG AI Research earned the top expert evaluation score of 31.6 out of 35. The five-team average was 28.56.
For the user evaluation, 49 AI expert users, including AI startup CEOs, participated. Based on AI-use websites built with the elite teams' AI models, they conducted an in-depth analysis of real-world applicability and inference cost efficiency. LG AI Research also earned the top user evaluation score of 25.0 out of 25. The average score was 20.76.
◇ Naver Cloud fails to meet originality requirements and is eliminated
The government also conducted an originality evaluation spanning △technical △policy △ethical perspectives.
On the technical side, the government aims for "independent implementation of the AI model," in which the entire training process is carried out through original AI model architecture design, independently securing and processing large-scale data, and applying proprietary training algorithm techniques. While leveraging open source is a general trend in the global AI ecosystem, the government views it as a basic condition for independent AI models, broadly recognized across domestic and international AI industry and academia, to develop AI models by initializing weights and then training. Therefore, even if elite teams strategically use proven open source to utilize validated technologies, align with the global AI ecosystem, and pursue global expansion, the government determined that training and development after initializing weights is the minimum requirement to secure a model's originality.
On the policy side, when using foreign AI models for national defense, diplomacy, security, or national infrastructure (power grid, transportation, communications), the goal is to mitigate potential risks such as the leakage of state secrets or threats to national security by securing the capacity to independently develop and advance AI models at any time (sovereignty) and to autonomously control the operation and use of AI models under any circumstances (control). In other words, AI models should be developed entirely with domestic technology or independently developed and advanced using open source without license restrictions, and they should be free from external control or interference resulting from open-source use, according to the government.
On the ethical side, given that using open source is standard in today's AI ecosystem, it is necessary to promote sound development by complying with license policies such as reference notices for AI models developed using open source, thereby enhancing trust in the AI ecosystem, strengthening open verification, and improving transparency.