LG Ships K-EXAONE 2.0, a 750B-Parameter MoE, Under Apache 2.0 — South Korea's Largest Model

The Korea Times·high signalXBlueskyLinkedInCopy link

LG AI Research released K-EXAONE 2.0 to Hugging Face on July 31 under South Korea's Sovereign AI Foundation Model Project: a hybrid-attention mixture-of-experts model with 750B total parameters, ~37B active per token, 256 experts with 8 selected per token, a 262,144-token context window, and 10-language coverage. It is more than 3x the size of the first K-EXAONE (236B total / 23B active) and lifts the average benchmark score from 63.3 to 70.1, with reported 83.5 on MMLU-Pro and 92.3 on AIME 2026. The weights are Apache 2.0, making this one of the largest permissively-licensed open-weight models available to builders.