The frontier-model race is no longer a two-horse contest. LG AI Research has published K-EXAONE 2.0, a 750-billion-parameter mixture-of-experts model that posts strong benchmark scores and signals South Korea’s growing ambitions in advanced AI.
The model reportedly scores 83.5 on MMLU-Pro and 92.3 on AIME 2026, placing it in competitive territory against leading systems on reasoning and knowledge tasks.
K-EXAONE 2.0 lands amid several defining trends in the field: reasoning-focused models that trade raw speed for accuracy, multimodal capabilities becoming standard, and rapid efficiency gains that deliver top-tier performance at falling costs.
For a market long dominated by a handful of US labs, LG’s release is a reminder that serious model development is spreading across regions and industries. As more players ship large, capable systems, the competitive pressure — and the pace of improvement — only intensifies.
Comments 0