The announcement of the second-round evaluation results for the 'Independent AI Foundation Model' project, which selects national AI (artificial intelligence) models, is imminent. With expectations that global benchmarks will carry greater weight in this round, attention focuses on how much the recently released 'Artificial Analysis Index' will influence the outcome.
According to the ICT industry on the 17th, the Ministry of Science and ICT plans to soon announce the results of the second-stage evaluation for the Independent AI Foundation Model project (hereinafter Dokpamo). The companies selected for this second-round evaluation are Motif Technologies (Motif), Upstage, SK Telecom, and LG AI Research.
Previously, the Ministry of Science and ICT established gender and age quotas based on resident registration population statistics to select 200 national evaluators. These evaluators directly used the four independent AI models from the 8th to the 11th and conducted absolute evaluations. The Ministry plans to announce the final evaluation results sometime next week, incorporating the national evaluators' assessments, expert consultations, and global benchmark evaluations. In this second round, three companies will be selected, and after a third-round evaluation at year-end, two companies will ultimately be designated as national AI models.
This evaluation is expected to focus on how much model scale and performance were improved while enhancing computational efficiency. The Dokpamo second-round evaluation indicators remain the same as the first round: 40 points for benchmarks, 35 points for expert evaluation, and 25 points for user evaluation. Given the significant controversy over the 'independence' criterion in the first round, this time the criteria have been clarified. Models must be independently developed from architecture design through training, not merely fine-tuned versions of foreign models.
The most significant variable this time is the global AI model evaluation by Artificial Analysis (AAII), a leading institutional investor in AI assessments. With AAII scores accounting for 25 out of the 40th benchmark points, Motif's model achieved the highest score of 47 among the four companies in the AAII evaluation released on the 13th, securing a favorable position. The second place went to Upstage's 'Solar Open2' with 37 points, followed by SKT's 'A.X-K2' at 35 points, and LG AI Research's 'K-Exaone 2.0' at 31 points.
AAII scores do not ultimately determine selection. However, since LG AI Research has maintained absolute leadership throughout the entire process since the Dokpamo project began, this result is considered somewhat surprising. In the first-round evaluation, LG AI Research ranked first across all evaluation indicators.
LG AI Research and SKT, which received relatively lower scores, have larger model sizes. 'K-Exaone 2.0' has 750B (750 billion) parameters, making it the largest among the four models. SKT's 'A.X K2' follows with 688B, 'Motif-3' with 314B, and 'Solar Open2' with 250B. Parameters are akin to an 'AI brain size'; the larger they are, the more data can be processed and complex patterns recognized.
Meanwhile, companies that did not advance from the first round of Dokpamo are gaining opportunities to participate in various national AI projects. Naver, which faced setbacks in the first round, was selected as one of the contractors for a 2 trillion won national AI GPU data center construction project, and Naver Works was also chosen as a government-based platform by the Ministry of Science and ICT and the Ministry of the Interior and Safety. NC AI was selected as the lead institutional investor for the government's agentic AI technology development project.
in TECH
