LG AI Research Releases K-EXAONE 2.0, South Korea's Largest AI Model
LG AI Research has released K-EXAONE 2.0, a Mixture of Experts (MoE) model with 750 billion total parameters and 37 billion active parameters. Released under the Apache 2.0 license, it is currently South Korea's largest AI foundation model. This release marks a major milestone for South Korea's sovereign AI development, offering a highly capable open-source model that competes with leading global models. Its open-source nature allows developers worldwide to leverage a massive-scale MoE architecture without proprietary restrictions. K-EXAONE 2.0 utilizes a hybrid attention MoE architecture and achieved an average score of 70.1 across 24 benchmarks, a 10% improvement over its predecessor. It outperforms models like GLM-5.1 and Qwen3.5 in long-context understanding and agent tool usage.
## BACKGROUND
A Mixture of Experts (MoE) architecture scales model capacity by routing inputs to specific subnetworks, known as experts, rather than activating all parameters for every token. This approach allows models to achieve high performance with significantly lower computational costs during inference.