K-Exaone 2.0

K-Exaone 2.0 is LG AI Research's sovereign AI flagship and, at 750 billion parameters, South Korea's largest independently trained AI model, unveiled around August 2026. LG said its researchers independently completed the architecture, training data, distributed computing, and inference infrastructure, marking the country's first successful large-scale training run at this scale.

The model supports ten languages, adding French, Italian, Portuguese, and Polish to Korean, English, Spanish, German, Japanese, and Vietnamese. LG positioned it strongly on alignment: it reportedly scored 94.6 on average in tests covering Korean social norms, global ethical standards, and geopolitical context, versus 71.3 for GLM-5.1, 65.2 for DeepSeek V4 Pro Max, and 89 for Qwen3.5, per company-reported figures.

K-Exaone 2.0 is LG's bid for sovereign AI credibility amid Korea's national push for domestic models tailored to local language, culture, and regulation. LG planned a public evaluation website and a consumer service so users could test the model directly. Exact release date is era-level; it surfaced in Korean press coverage in August 2026.