LG unveils K-Exaone 2.0, Korea’s largest open-source AI model

LG AI Research unveiled AI model K-Exaone 2.0 on the open-source platform Hugging Face, the company said Friday.

K-Exaone 2.0 is the second model under the Ministry of Science and ICT's initiative to develop a domestic sovereign AI platform. It is the largest AI foundation model built in Korea to date, according to the company. LG AI Research is a corporate institute dedicated to LG Group’s AI ambitions.

K-Exaone 2.0 has 750 billion parameters. It tripled the size of the project's first model, which had 236 billion parameters. Parameters are the internal values an AI model adjusts during training. The more parameters a model has, the better it can represent information and perform reasoning tasks.

LG AI Research also adopted the Apache 2.0 license to allow anyone to use the model commercially and expand its accessibility.

The institute also released the results of its performance evaluation across nine categories and 24 benchmarks. K-Exaone 2.0 posted an average score of 70.1 across the 24 benchmarks, up more than 10 percent from the first model's score of 63.3.

Performance on three key coding and agentic coding benchmarks improved by about 30 percent.

The model scored 94.4 on OpenAI-MRCR, which measures long-context comprehension in English, and 89.6 on the Korean-language benchmark Ko-LongBench. Those scores exceeded Chinese Zhipu AI's GLM-5.1, which recorded 71.5 in English and 83.6 in Korean.

K-Exaone 2.0 also matched or outperformed leading global AI models, such as GLM-5.1, DeepSeek V4 Pro Max and Qwen3.5 on benchmarks evaluating agentic tool use and following instructions — tested through Tau3-Bench Banking, IFEval and IFBench.

"K-Exaone 2.0  represents far more than simply building a larger model," Lim Woo-hyung, co-president of LG AI Research, said. “Our research team independently completed every stage of development — from model architecture and data preparation to large-scale distributed training and inference infrastructure — for the 750-billion-parameter foundation model.”

Lim noted that the development proved LG’s capability to compete alongside global frontier models.

“We will continue advancing K-Exaone through higher-quality data, continued post-training, reinforcement learning and more sophisticated inference technologies,” Lim said.

BY KIM KYUNG-MI [[email protected]]