[CLSTR]
< Back to situation

[REVISION HISTORY]

South Korean AI foundation model project development

Updated 1 time since CLSTR started tracking revisions of this situation.

What changed

2026-08-24 06:31 UTC → 2026-08-27 01:58 UTC · added removed

South Korea is advancing its national AI foundation model project, known as the ‘Dokpamo’ project, through significant infrastructure investment and competitive selection phases. Initially, the Ministry of Science and ICT moved to the third stage of the project by selecting Upstage, SK Telecom, and LG AI Research. To support these teams, the government is allocating approximately 3,000 NVIDIA B200 GPUs, with 1,000 units assigned to each selected group. This follows broader government efforts to secure massive GPU infrastructure and develop technologies for flexible resource sharing. However, the selection process has faced scrutiny regarding evaluation transparency. While the three aforementioned teams advanced, Motif Technologies was eliminated despite reportedly achieving has filed a formal objection regarding its exclusion from the highest score second stage of the project. Despite its ‘Motif3’ model ranking first among four elite teams in the global ‘Intelligence Index’ benchmark by Artificial Analysis Intelligence Index (AAII) benchmark. Critics have raised concerns because Analysis, the ministry has not disclosed total company received low scores or comprehensive rankings, making it difficult to verify why a top-performing model was excluded. In response, in expert and user evaluations during the National Information Society Agency (NIA) government’s assessment. Motif is working to expand demanding that the Ministry of Science and ICT disclose detailed evaluation criteria to include tool use, agents, scores, specific criteria, and coding. Industry experts are also advocating the methodology used for more diverse global indicators user and blind testing to prevent models from being optimized evaluations. The company has specifically for known datasets. questioned the scoring conversion process, noting that a significant 16-point gap in global benchmark scores was reduced to only a 4-point difference when converted to the government’s 25-point scale. Motif stated its goal is to seek transparency and improvements in the evaluation process rather than simply demanding selection.

Versions

  1. 2026-08-27 01:58 UTC South Korean AI foundation model project development
  2. 2026-08-24 06:31 UTC South Korean AI foundation model project development

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.