- Public–private collaboration to develop a globally influential Korean AI foundation model

- Up to five elite teams will be selected, aiming for ≥95% performance against leading global models

- Comprehensive support (GPU resources, data, talent) for up to three years per team

The Ministry of Science and ICT (Minister Yoo Sang‑im, hereinafter “MSIT”), alongside the National IT Industry Promotion Agency (President Park Yun‑kyu, “NIPA”), the National Information Society Agency (President Hwang Jong‑sung, “NIA”), and the Institute for Information & Communications Technology Planning & Evaluation (President Hong Jin‑bae, “IITP”), has launched an open call (June 20–July 21) for elite domestic teams to participate in the “Proprietary AI Foundation Model”* project.

* The initiative was previously referred to as the “World Best LLM (WBL)” project.

Background

With AI supremacy becoming a matter of national importance worldwide, securing a globally competitive proprietary AI foundation model has become ever more critical. By making such a model available as open source in Korea, this project aims to accelerate AI-driven service launches and broader societal innovation.

Building on the directions set out in the National AI Capability Enhancement Plan (released in February 2025), the government has prepared this public–private collaboration to recruit top-tier domestic teams for model development.

Elite Team Recruitment

 Target applicants: High-performing Korean AI companies and institutions (consortium applications accepted)

 Selection method: Up to five teams selected; phased competitive rounds (e.g., 5 → 4 → 3 → 2)

 Requirement: Participation by university or graduate students is mandatory to foster future AI talent

 Performance goal: Achieve ≥95% benchmark performance relative to global models released within the past six months

 Approach: Teams must propose their own development strategies and methodologies. The evaluation criteria will adopt a flexible “moving-target” approach, allowing adjustments at each stage to reflect rapid technological changes—encouraging a DARPA-style competitive development model.

Teams may freely propose the resources required for development—including GPUs, data, and talent—based on their own plans. Final allocations will be determined through the evaluation process.

Teams are also required to submit plans outlining how their models will improve public access to AI and support AI adoption across public, economic, and social sectors. These plans will be considered in the evaluation.

Matching ratios will be adjusted according to each team’s size and level of open-source commitment. Teams that offer broader open-source access will be eligible for a lower cost-sharing requirement.

Evaluation & Selection

An expert panel featuring domestic and international AI specialists will conduct evaluations.

For the six-month phase evaluations (starting in December), criteria include:

1. Contest-style evaluation by a jury of general public and experts

2. Benchmark-based evaluation using international and domestic standards, combined with Korean-language performance and safety validation systems

3. Derivative-model evaluation (based on the number and quality of derivative models)

Details of the phased evaluation framework will be finalized in consultation with the selected teams.

Support for Model Development

Comprehensive support will be provided across three categories—GPUs, data, and talent:

1. GPU support:

 Until mid‑2026: Rent privately owned GPUs using the KRW 157.6 billion supplementary budget

 Afterwards: Distribute up to 10,000 GPUs purchased by the government

 Initially, approximately 500 GPUs will be allocated per team, with the number increasing to over 1,000 based on evaluation results.

※ Companies that have leased GPUs to the government are ineligible for support from government-purchased GPU stock until after mid-2026.

2. Data support:

 Teams can participate in a joint dataset purchase (up to KRW 10 billion/year per team; amount may vary with phased team selection).

 An additional KRW 3–5 billion/year per team for proprietary data generation and preprocessing (also subject to phased selection).

3. Talent support:

 Teams that recruit overseas or Korean diaspora researchers will receive government matching support for salaries and research expenses (approximately KRW 2 billion per year).

 Talent support will continue through 2027, regardless of team down-selection status.

※ Support scales remain subject to adjustment based on evolving conditions.

Follow‑Up and Dissemination

Models developed under this project will be open‑source and integrated into Korea’s AI ecosystem. Participating teams will be required to present detailed plans to enhance public access to AI and support the adoption of AI across public, economic, and social domains. These plans will be considered in the selection process, and the government may co-support implementation through relevant government-funded programs. In collaboration with the AI Safety Institute, the government will also ensure that the foundation models undergo rigorous safety validation. Additionally, official recognition such as the “K-AI Model” or “K-AI Company” label may be granted to strengthen global credibility and accelerate international market entry.

Official Statement

Deputy Minister Song Sang-hoon of the Office of ICT Policy at MSIT stated:

“This project goes beyond simply developing a proprietary AI foundation model—it aims to secure Korea’s technological sovereignty in the AI era and build an ecosystem that enables shared growth.

We expect elite domestic teams to rise to the challenge, secure globally competitive foundation models, and pave the way for Korea to thrive as an AI powerhouse by ensuring nationwide public access to AI technologies.”

For further information, please contact the Public Relations Division (Phone: +82-44-202-4034, E-mail: msitmedia@korea.kr) of the Ministry of Science and ICT. 

Please refer to the attached PDF.