South Korean AI startup Motif Technologies has released its self-developed large language model "Motif 3" as open source, recording the highest score among participants in the country's sovereign AI foundation model (Dokpamo) initiative in a global AI performance evaluation. Ahead of the second-phase evaluation of the government-led Dokpamo program, the late entrant is being viewed as having pulled off an upset over major corporations and established players. Motif Technologies announced on the 13th that it had released Motif 3 on the AI model platform Hugging Face. Motif 3 is a Mixture of Experts (MoE) architecture model with 314 billion (314B) total parameters, of which only 13.2 billion (13.2B) are active parameters that actually participate in computation. The company said it developed the entire process — from pre-training to post-processing — from scratch in approximately five months. In the latest Artificial Analysis Intelligence Index (AAII) from global AI evaluation firm Artificial Analysis, Motif 3 scored 47 points. This is the highest score among the four models participating in the second-phase Dokpamo evaluation. Upstage's "Solar Open 2" followed with 37 points, SK Telecom's "A.X-K2" with 35 points, and LG AI Research's "K-EXAONE 2.0" with 31 points. Motif 3 opened up a 10-point gap over second place and a