LG AI Research Releases K-EXAONE 2.0 750B A37B Open-Weights MoE Model
LG AI Research has released K-EXAONE 2.0 750B A37B, a 750-billion parameter Mixture-of-Experts (MoE) model with 37 billion active parameters. The model is licensed under Apache 2.0 and optimized for 10 languages, including Korean, English, and Vietnamese. This release represents a major open-weights contribution to the global AI ecosystem, supporting sovereign AI initiatives and offering high-performance multilingual capabilities. It provides a powerful, open alternative to proprietary models for complex tasks like agentic tool use and long-context reasoning. Developed under Phase 2 of Korea's Sovereign AI Foundation Model Project, the model is three times larger than its predecessor. It achieves strong benchmark scores, including a 94.6 average on safety benchmarks like ROK-Fortress and outperforming GLM-5.1 in long-context tasks.
## BACKGROUND
The Korea Sovereign AI Foundation Model Project is a national initiative aimed at building independent AI capabilities to reduce reliance on foreign technology. Mixture-of-Experts (MoE) is an AI architecture that activates only a subset of its total parameters (in this case, 37 billion out of 750 billion) for any given input, improving computational efficiency.