South Korea's Ministry of Science and ICT released additional detailed results from the second-stage evaluation of its independent AI foundation model project.

The ministry said on Wednesday that Motif Technologies received the highest score in a global benchmark assessment. SK Telecom received the top score in the Korea Intelligent Information Society Agency (NIA) benchmark and the AI expert user evaluation. LG AI Research recorded the highest score in the expert evaluation and the general public evaluation.

The ministry had announced the results of the second-stage evaluation on Aug. 18 but did not disclose the top-ranked company and score for each category. It explained that it considered criticism that disclosing category winners in the first-stage evaluation could give the companies an excessive boost in recognition and create a stigma effect for the others.

It expanded the scope of disclosure after transparency concerns were raised. The second-stage evaluation comprised 40 points for benchmark assessment, 35 points for expert assessment and 25 points for user assessment.

In the benchmark assessment, the global review using the Artificial Analysis Intelligence Index (AAII) gave Motif Technologies the highest score, at 11.9 out of 25. The four-team average was 9.48.

Global benchmarks used in the AAII assessment included GDPval-AA v2 and tau3-Banking in the agent category, Terminal-Bench v2.1 and SciCode in the coding category, and general and scientific reasoning benchmarks. The agent category carried the largest weight at 34 percent of the overall AAII assessment.

In the NIA benchmark, SK Telecom received the highest score, at 13.4 out of 15. It assessed seven areas including mathematics, knowledge, long-context comprehension, safety, reliability, Korean language and instruction following, and the four-team average was 13.05.

In the expert assessment, worth 35 points, LG AI Research recorded the highest score at 29.5. The four-team average was 28.75.

The expert assessment was conducted by external experts, including three from industry, five from academia and two from the research community, who evaluated development strategy and technology, development results and plans, and ecosystem and global ripple effects. Evaluation items also included model usability and practicality and applicability across various domains and scenarios.

The user assessment comprised 15 points for an AI expert user evaluation and 10 points for a general public evaluation. The AI expert user evaluation included 49 participants such as AI startup CEOs, and SK Telecom received the top score at 11.6. The four-team average was 10.53.

In the general public evaluation, 185 of 200 randomly selected people participated, using gender and age quotas. LG AI Research received the highest score at 7.6 out of 10, and the four-team average was 7.03.