Korea's national AI teams complete independent models as second evaluation approaches

Second evaluation set for early August; field narrows from four to three teams

'Practical applicability' expected to be key criterion

Elite teams racing to build real-world use cases

Competition among South Korea's elite AI teams is intensifying as the second evaluation round of the Independent AI Foundation Model Project — a government initiative to select a national AI champion — draws within a month.

LG AI Research, SK Telecom and Upstage have all finished developing their independent models. Motif Technologies, which joined the project through a supplementary open call, is on track to complete development by the end of this month.

The second evaluation, scheduled for early next month, will narrow the field from four teams to three, raising the question of whether South Korea can build a solid foundation for sovereign AI in a race that has become a global sprint.

According to industry sources, LG AI Research, SK Telecom and Upstage completed development of their independent AI models at the end of June and are set to submit performance reports to the National IT Industry Promotion Agency (NIPA) in the coming days.

Motif Technologies, which joined the project in February, has been guaranteed the same development window as the other teams and will complete its model by the end of this month.

File photo of AI technology. [Getty Images Bank]

The government plans to hold the second evaluation in early August, once all four teams have finished model development, and will eliminate one team at that stage. The final two teams are scheduled to be confirmed in February next year. The original plan called for selecting the final two teams by December this year, but the timeline slipped somewhat after an additional open call and related procedures followed the first evaluation.

Industry observers expect the second evaluation to assess "proven technology" — AI that can be deployed directly in real-world settings. While the broad framework from the first evaluation will be maintained — benchmark assessment, expert review and user evaluation — practical applicability in actual industrial environments is expected to emerge as the central criterion.

In line with that expectation, the elite teams are pouring resources into expanding real-world use cases for their AI technology.

LG AI Research's K-Exaone has been building its track record after its technology was recently integrated into the portal Zum's AI search service. Zum replaced the core AI model powering its "AI 3-Second Summary" and "AI Issue Trends" services with K-Exaone.

K-Exaone is already deployed in the production and manufacturing processes of LG Group's manufacturing affiliates, further expanding its ecosystem.

A visitor tries SK Telecom's A.X K1 model at the Independent AI Foundation Model Project's first showcase, held at Coex in Gangnam-gu, Seoul. [Provided by SK Telecom]

SK Telecom is pursuing field demonstrations of its independent AI foundation model with steelmaker KG Steel and auto-parts manufacturer Konex. The company has developed a demo version of a "manufacturing-specialized AI agent" based on its A.X K1 model and plans to deploy it in actual production processes in the second half of this year.

Upstage is pursuing a strategy centered on its Solar WBL model, targeting the finance and manufacturing sectors. The company is also rolling out Solar across Daum, the portal it acquired, positioning the model as an AI for the general public.

Motif Technologies, the last team to join the elite lineup, is advancing its technology on the foundation of its in-house large language model Motif-2-12.7B. The company is developing a 300-billion-parameter LLM for the second evaluation and aims to push beyond language models to build capability in image and video generation as well.

Meanwhile, the Ministry of Science and ICT does not plan to hold a public showcase ahead of the second evaluation, as it did before the first. The ministry has also shared with the teams the refined sub-criteria for the second evaluation that were adjusted following the first round.

sjpark@heraldcorp.com