South Korea's Largest AI Foundation Model: LG AI Research Institute Releases 750B Parameter K-EXAONE 2.0

LG AI Research Institute has officially launched K-EXAONE 2.0, the largest-scale AI foundation model developed domestically in South Korea. Based on a hybrid attention MoE architecture, the model boasts 750B total parameters, more than tripling the scale of the first-generation model.

While maintaining a massive total parameter count, the new model keeps active parameters at 37B. This design helps balance inference efficiency with model capability. The model is open-sourced under the Apache 2.0 license, allowing the community to freely use and modify it.

As a project outcome supported by the Korea Sovereign AI Foundation, the release of K-EXAONE 2.0 marks a further enhancement of South Korea's technical strength in the large model field. Its open-source strategy is expected to lower the barrier for local Korean enterprises and application developers to access large model capabilities.

In 24 benchmark tests, the model demonstrated strong average scoring capabilities. As global competition in large models intensifies, South Korea aims to secure a more favorable position in the Asian AI ecosystem through such open-source foundation models.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.