South Korea's sovereign AI model competition enters its final stretch as four teams — LG AI Research...
The AMW Read
Incremental update to known players in Korea's sovereign AI push, but the multi-team evaluation with strategic differentiation has segment-level significance.
Named counterparties: Motif
South Korea's sovereign AI model competition enters its final stretch as four teams — LG AI Research Institute, SK Telecom (017670.KS), Upstage, and Motif Technologies — proceed to the second-stage evaluation. The Ministry of Science and ICT confirmed that Epoch AI, a global nonprofit research institute, designated the latest models from all four teams as "Notable AI Models," based on criteria including technical influence and research significance. LG's K-EXAONE 2.0 leads with 750 billion total parameters (37 billion activated under a mixture-of-experts architecture), while SK Telecom's A.X K2 expands to 688 billion parameters. Upstage's Solar Open2 uses 250 billion total parameters (15 billion activated) with a hybrid attention structure supporting up to 1 million token contexts, and Motif's model also made the list.
The strategic divergence among the four teams mirrors the broader global shift from chatbots to AI agents. LG emphasizes general problem-solving, scoring 92.3 on AIME 2026 math benchmarks and improving 30% in coding and agent-related assessments. SK Telecom is betting on a "full-stack" approach that extends from the model to enterprise AI services and AI data centers (AIDC), with use cases at KG Steel, Konec, the Ministry of National Defense, and SK Biopharmaceuticals. Upstage is positioning Solar Open2 specifically for AI agents that handle long-running, multi-step tasks, leveraging its long-context design for efficiency. Motif rounds out the field as a challenger.
For builders and investors, the competitive dynamics hinge on how government evaluation weights benchmark scores against real-world industrial deployment. LG's top finish in the first round faces pressure from SKT's infrastructure-linked strategy and Upstage's agent-first efficiency play, as reflected in the AI Market Watch index tracking Upstage's trajectory over the past 90 days alongside its prior launches. Startups like Upstage and Motif demonstrate that parameter count alone doesn't determine competitive advantage — efficiency, context handling, and vertical integration matter equally. The selection outcome will shape Korea's sovereign AI stack and set a precedent for how national champions balance open-weight innovation with enterprise readiness.
