One More Elite Team to Be Recruited for 'Independent AI Foundation Model' Project... 'Originality' Evaluation Criteria Strengthened

Applications Open Until February 12
Second Phase Evaluation Scheduled for Early August
Support Equivalent to 768 B200 GPUs Provided

The government will recruit one additional elite team (consortiums allowed) for the "Independent AI Foundation Model" project. The second phase, originally designed as a competition among four elite teams, began with only three teams after both the Naver Cloud and NC AI consortiums were eliminated in the first round of evaluations. To fill this vacancy, the government intends to select another elite team. The evaluation will focus on technical capabilities that enable meaningful competition with the existing elite teams and the ability to contribute to the growth and expansion of the domestic artificial intelligence (AI) ecosystem.

On January 23, the Ministry of Science and ICT announced, together with the National IT Industry Promotion Agency (NIPA) and the National Information Society Agency (NIA), the launch of an additional call for elite teams for the Independent AI Foundation Model project. The application period runs from today until February 12.

This additional recruitment is also related to the recent controversy over originality that arose after the results of the first phase evaluation. Contrary to expectations that only one team would be eliminated, Naver Cloud failed to advance to the second phase due to controversy over "from scratch" development (designing the model entirely from the ground up, rather than modifying an external model). This led to ongoing discussions about the interpretation and application of evaluation criteria.

Strengthening the "Originality" Criterion in Expert Evaluation

To ensure the newly selected team is given the same AI model development period as the existing elite teams, the government plans to conduct the phase evaluation around early August. The evaluation process will remain the same as before, consisting of: ▲ benchmark evaluation ▲ expert evaluation ▲ user evaluation. The government is also considering applying benchmarks targeting major global leaderboards. In addition, there are plans to strengthen the "originality" criterion in the expert evaluation category.

As with the initial call, the additional recruitment targets elite teams centered on domestic AI companies and institutions. Participating teams must present a concrete model development strategy and methodology, aiming for performance at least 95% that of the latest global AI models.

In particular, the one additional elite team to be selected must be able to compete meaningfully with the existing three elite teams in terms of technology, and must possess capabilities that go beyond simple AI model development to contribute to the growth and expansion of the domestic AI ecosystem. The government will conduct a comprehensive evaluation of various indicators and AI model development capabilities, and will select a team recognized by a majority of expert evaluators. If no suitable team is identified through the evaluation, the government may choose not to select an additional team.

The selected additional elite team will receive support equivalent to that provided to the existing elite teams. The government will provide GPU resources equivalent to 768 B200 units (comparable in performance to 512 B200 units and 512 H100 units), as well as support for joint data purchases and data construction and processing. The selected team will also be granted the "K-AI Enterprise" designation. However, companies that have leased GPUs to the government will be excluded from GPU support for this project, and will only be eligible for support starting from the point when purchased GPUs are utilized (in the second half of 2026).