Motif Technologies

AI · Large Language Models · South Korea

Milestone

Motif's Motif 3 tops Korean AI models, ranks 9th worldwide

The five-month-old model beat two conglomerate rivals in a contest to pick Korea's sovereign AI champion.

• Motif Technologies' "Motif 3" scored 47 on the Artificial Analysis Intelligence Index (AAII), the top mark among four models competing in the second round of Korea's government-backed sovereign AI foundation model project, Dokpamo.
• Motif Technologies is a Korean AI startup that built Motif 3, a 314-billion-parameter mixture-of-experts model, entirely from scratch—pretraining through post-processing—in about five months, then released it open source under an MIT license.
• Motif 3's 47 beat Upstage's Solar Open2 (37), SK Telecom's A.X-K2 (35), and LG AI Research's K-EXAONE 2.0 (31)—notable since LG had led the first Dokpamo round.
• The model ranks 9th globally among all LLMs and 4th among open-weight models on AAII, yet activates only 13.2 billion of its 314 billion parameters at inference, versus rival open-weight models running near 1 trillion parameters.
• AAII is worth just 25 of 100 total evaluation points in Dokpamo's second round, with a national-agency benchmark, expert review, and user testing still to be tallied before one of the four teams is cut.
• A young, comparatively resource-poor startup outscoring two of Korea's largest tech conglomerates on a global benchmark upends assumptions about who is best positioned to lead the country's state-backed AI push.
• Motif's CEO says the next model, Motif 4, targets top-5 global performance and plans to fuse the vision, audio, and video encoders released alongside Motif 3 into one multimodal system.