Gov’t to conduct 1st evaluation of national AI foundation model project in January

Minister of Science and ICT Bae Kyung-hoon speaks during a press conference announcing the artificial intelligence policy road map for science and technology in Seoul, Sept. 12. Courtesy of Ministry of Science and ICT

The Ministry of Science and ICT will carry out the first round of evaluations for the state-led project to develop homegrown artificial intelligence (AI) foundation models in January.

The ministry notified the participating consortia on Thursday that performance assessments to eliminate one of the five teams will take place from Jan. 1 to 15.

In August, the ministry selected five consortia led by Naver Cloud, Upstage, SK Telecom, NC AI and LG AI Research to locally develop foundation models, investing around 200 billion won ($136 million) to cover costs for high-performance GPUs, datasets and R&D personnel.

Every six months, the government will eliminate one of the teams after evaluation, narrowing the field to two finalists by the first half of 2027. The upcoming January assessment will determine the first team to be eliminated.

The evaluation was initially slated for December but was pushed back to accommodate delays from issues running graphics processing units (GPUs).

Some of the participating companies reportedly faced problems getting the GPUs provided by SK Telecom and Naver Cloud to run to full performance specifications, which slowed their development progress.

The ministry explained that such technical issues are a normal part of building large AI models, especially in early stages.

“When developing large-scale AI models using massive GPU clusters, it is essential to solve certain challenges at the beginning, such as getting the GPUs set up, making sure all the systems work well together and fixing any technical issues that come up during the process,” the ministry said.

“The government has been actively helping the selected project teams utilize GPUs from the very start and is closely coordinating with all teams to address technical issues as they arise.”

Each consortium is aiming to build a model that matches at least 95 percent of the performance levels of top global models. At each evaluation, the teams’ technology, potential industry impact and plans for global growth are considered.