Invest1 publisher2 min readPublished
Artificial Analysis's Intelligence Index carries a quarter of Korea's sovereign AI score
Korea's Dokpamo program gives the benchmarking firm's index 25 of the 100 points that decide which teams advance. In the results published on August 27, the team that led that component finished last of the four.
The Investor · Invest desk
What happened
- Artificial Analysis is the official global evaluation partner for South Korea's Sovereign AI Foundation Model project, and its Intelligence Index accounts for 25 of the 100 points in second-phase scoring.
- Second-phase results published on August 27, 2026 put SK Telecom first on 70.6 points, Upstage on 69.9, LG AI Research on 69 and Motif Technologies on 65.8.
- Motif Technologies posted the best score of the four teams on the Artificial Analysis component, 11.9 points, and still finished last on the combined scorecard.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint Tuning for the public index has a ceiling: the best performance on it so far bought 11.9 points. That limits how much of a 100-point selection an external evaluator can swing.
- exposure Korean teams chasing state compute are now scored in part by a private firm whose methodology they do not control and which can be revised between phases.
- decision MSIT and NIPA have to settle before phase three whether a quarter of the score stays with an outside index after a round in which that index's leader placed fourth.
Motif Technologies led the Artificial Analysis component with 11.9 points and finished fourth of four [5][4]. The component was worth 25 [2]. On the published figures nobody turned more than 11.9 of those 25 into score. That puts 53.9 of Motif's 65.8 on the other 75 points, and at least 58.7 of SK Telecom's 70.6 in the same place [12][13]. The team that won the outside benchmark lost the round by 4.8 [13].
SK Telecom's margin was 0.7 over Upstage and 1.6 over LG AI Research [14]. The report publishes Motif's index sub-score and does not publish the other three teams', so nothing in the phase-two results shows whether the component separated the top two [17].
What the four teams are competing for is compute. Each reportedly gets access to about 1,000 NVIDIA B200 GPUs for six months, plus proprietary datasets and direct funding [8]. Four teams came through a first-round cut from five, so roughly 4,000 B200s are under allocation [6][15].
The Ministry of Science and ICT and the National IT Industry Promotion Agency launched Dokpamo in 2025 to build pre-trained models Korea fully controls [7]. The logic was that relying on American or Chinese foundation models means accepting their licensing terms, export controls and strategic priorities [7]. A quarter of the score that decides which Korean teams advance now moves with a methodology Artificial Analysis maintains [2][1]. Cryptobriefing, which reported the partnership, wrote that the Intelligence Index "synthesizes performance across multiple dimensions, giving a composite score that's harder to game than any single benchmark" [9].
The weight can hold, and teams train toward the index, in which case phase-three sub-scores climb toward the 25-point ceiling. MSIT can trim the weight after a round in which the index leader placed last. Or the weight holds, the top of the table stays inside a point, and the other 75 points keep deciding placement. I would take the third: a 0.7-point race gets decided on the other 75, not on a component whose best observed yield is 11.9 out of 25 [5][2][14]. The evidence here is one program's scoring sheet. If Artificial Analysis revises the index between phases and that moves a team three or four points, the 0.7 at the top becomes the component's to decide, and I would be wrong.
What to watch
- Phase-three criteria from MSIT and NIPA, and whether the Artificial Analysis component keeps its 25 of 100 points.
- Publication of Intelligence Index sub-scores for all four teams. Those would show whether the component separated SK Telecom from Upstage.
- A second national program naming Artificial Analysis in its scoring rules. That would be the evidence that another government has adopted the same weighting.