The Ministry of Science and ICT is considering a major restructuring of its 'Independent AI Foundation Model' (Dokpamo) project, which selects national representative AI companies. This comes from the assessment that the performance gap between domestic AI models and global frontier AI models is significant.
On the 18th, the Ministry of Science and ICT and the Korea Information Society Development Institute (NIPA) held a briefing on the second-stage evaluation of the Dokpamo project, announcing that three teams — Upstage, SK Telecom, and LG AI Research — have advanced to the next stage. Motive Technology, which joined as the new national representative through a 'relegation playoff,' unfortunately did not make it.
The second-stage evaluation of Dokpamo consisted of benchmark assessment (40 points), expert evaluation (35 points), and user evaluation (25 points). Among these, user evaluation had the greatest impact on determining which teams advanced or were eliminated.
In the user evaluation, composed of AI professional user assessment (15 points) and general public assessment (10 points), the four finalist teams received an average score of 17.6 points, with a 5.0-point gap between first and fourth place — the largest among all categories. The benchmark scores averaged 22.5 points, with a 4.0-point gap between first and fourth. Expert evaluation averaged 28.8 points, with a 2.4-point gap between first and fourth.
The user evaluation that determined advancement involved 49 AI professional users and 184 general citizens who comprehensively assessed the usability and perceived effectiveness of the AI models. A public evaluation panel (200 selected, 185 responded), which was not present in the first round, was added this time.
As a result, Upstage's finalist team received high scores for implementing a sovereign AI full stack by integrating with portals such as Naver and Timely platforms and using domestic FuriOSAI NPUs (Neural Processing Units).
SKT's finalist team demonstrated high performance in mathematical reasoning and Korean language domains, and further proved the potential of manufacturing-industry-specific models by actually providing an A.X K1 model for defense applications.
LG AI Research received meaningful evaluation for its ongoing collaboration with global international organizations, its differentiation in agentic AI, and its clear consideration of developing safe and trustworthy AI.
In Motive's case, although it achieved the highest score among the four companies (47 points) at AAII and demonstrated independence by developing its own architecture, tokenizer, optimizer, and kernel, it appears to have received relatively low scores in usability evaluation.
Ryu Je-myeong, Deputy Minister of Science and ICT, stated, "No single company ranked first across all categories," adding that "three companies were selected with only slight differences."
On this day, the Ministry of Science and ICT hinted at the possibility of restructuring the current Dokpamo project approach. With global frontier models advancing rapidly, it is considering shifting from distributing limited budgets for GPUs (Graphics Processing Units) among multiple companies to concentrating support on a single entity.
Ryu (Vice Minister) said, "There is growing consensus on moving away from the current competitive format and restructuring toward models comparable to frontier models," adding, "Competition surrounding AI models has intensified recently, and major leading countries are increasingly realizing moves to treat AI models as national security assets."
In fact, according to the AAII evaluation results released on the 13th, even Motive, which ranked first at that time, scored only 47 points — a significant gap compared to Claude Opus (63 points), Fable5 (62 points), and GPT 5.6 (61 points). If the original goal of Dokpamo — to reach 95% of global frontier model performance — is to be met, scores above 57 would have been required.
However, the Ministry of Science and ICT plans to proceed as scheduled with the third-stage evaluation, which will narrow the three selected companies down to two.
A ministry official stated, "When reviewing the development of frontier-level AI models, we plan to also examine the connectivity and restructuring with existing Dokpamo projects, AGI projects, and other related initiatives," adding, "We plan to announce the direction for frontier-level model development projects, including whether to continue the four-stage Dokpamo process, in the near future."
Meanwhile, the three companies advancing to the third-stage evaluation announced their goal to "challenge global standards." Motive, which was eliminated, stated, "We have proven that it is possible to develop AI models at the global frontier level even in Korea based solely on technological capability, and we will continue to dedicate our efforts to global model development in the future."
in TECH
