LG AI Research has unveiled K-EXAONE 2.0, the largest artificial intelligence (AI) model in South Korea, accelerating its push into the global AI market.
On Friday, LG AI Research released K-EXAONE 2.0 on the open-source platform Hugging Face. This is the second model in the Independent AI Foundation Model Project led by the Ministry of Science and Information and Communications Technology (ICT).
K-EXAONE 2.0 boasts 750 billion parameters, more than triple the size of its predecessor’s 236 billion. The model now uses the Apache 2.0 license, allowing for broader commercial applications.
LG AI Research reports that K-EXAONE 2.0 scored an average of 70.1 across 24 benchmarks in nine areas, including performance metrics for independent AI foundation models. This marks a 10% improvement over the first model’s 63.3 score. Notably, performance in key coding and agentic coding indicators improved by about 30%.
In critical areas like long-context understanding, agentic AI, and instruction following, K-EXAONE 2.0 matched or outperformed leading global models. It scored 94.4 on OpenAI-MRCR and 89.6 on Ko-LongBench for long-context understanding, surpassing China’s GLM-5.1. The model also showed superior performance in agentic tool use compared to GLM-5.1 and Qwen3.5.
The model now supports ten languages, adding French, Italian, Portuguese, and Polish to its existing repertoire of Korean, English, Spanish, German, Japanese, and Vietnamese.
In stability assessments, K-EXAONE 2.0 achieved an average score of 94.6 across key indicators evaluating adherence to Korean cultural nuances, global ethical standards, and geopolitical context judgment. This surpasses the stability of GLM-5.1, DeepSeek V4 Pro Max, and Qwen3.5.
LG AI Research independently managed the entire development process, from design to data training, distributed learning, and inference environment setup. This achievement positions them to compete with global frontier AI models.
This release is a crucial part of the government’s Independent AI Foundation Model Project, aimed at reducing reliance on foreign tech giants and fostering the domestic AI ecosystem.
Im Woo-hyung, co-director of LG AI Research, emphasized that K-EXAONE 2.0’s significance lies not just in its increased parameters, but in the fact that the domestic research team independently completed the entire development process for a 750B model. He pledged to further enhance global competitiveness by refining data quality, subsequent learning, reinforcement learning, and inference technologies.
Following the successful application of EXAONE 4.5 in manufacturing, biotech, finance, and public sectors, LG AI Research plans to unveil a new industry-specific AI foundation model next week. They are also developing a service that will allow the general public to interact directly with K-EXAONE 2.0.