The government on Aug. 18 released the results of the second evaluation of the homegrown AI foundation model project and said it will seek to overhaul the support format from the third evaluation.

According to the government announcement, three teams—Upstage, SK Telecom and LG AI Research—passed the second evaluation and advanced to the third stage.

Under the original plan, the government will select two final companies after the third evaluation. The companies will work in 2027 to develop AI models with performance at 95 percent of top global models.

Ryu Je-myung (류제명), second vice minister of the Ministry of Science and ICT, said at a briefing on the results that global frontier companies' model performance is advancing exponentially. He said he has been discussing with the selected teams whether they can keep up with global trends while maintaining the current scale of support and competitive structure, and signalled there will be changes in the third evaluation.

Ryu said it is difficult to disclose details because the government budget has not been finalised. He added there is a consensus to move away from the current competition format and reorganise the competitive structure so that models comparable to those of frontier companies can be built.

The ministry said the government is also discussing a budget for a frontier AI programme separate from the project. Ryu said he will explain again after the budget is finalised how the third evaluation, scheduled for the second half of the year, and a new competitive structure planned for 2027 will be linked.

On the detailed format of the third evaluation, Ryu said he will review what was learned through the first and second evaluations and check what needs to be supplemented. He said he will finalise the format through consultations with the three participating companies and announce it as soon as possible.

After the briefing, the ministry officially announced its position that the third round will proceed as originally planned.

The ministry said the government is reviewing the development of frontier-class AI models to respond to accelerating global AI competition. It said it plans to also review linkages and reorganisation with related programmes such as the existing project and the AGI project as part of that review.

The ministry plans to soon announce the direction of the frontier-class model development programme, including whether to continue the fourth stage of the project.