[DigitalToday reporter Chi-gyu Hwang] LG AI Research on Thursday released K-EXAONE 2.0, the second-round model under the Ministry of Science and ICT-led “Independent AI Foundation Model Project,” on the open-source community Hugging Face.

The institute said K-EXAONE 2.0 is a 750B (750 billion) parameter model, more than three times larger than the first-round model (236B).

LG AI Research stressed it also increased openness by switching the K-EXAONE 2.0 model licence policy to Apache 2.0, allowing anyone to use it commercially.

LG AI Research also released results from evaluations across 24 benchmarks in nine areas, including items reflected in performance assessments for independent AI foundation models.

K-EXAONE 2.0 scored an average of 70.1 points across the 24 metrics, showing performance more than 10 percent higher than the first-round model (63.3 points). The average performance of three key indicators in coding and agentic coding rose 30 percent.

It recorded performance that was higher or comparable to global leading models in key areas including long-context comprehension evaluation, agentic capabilities and instruction following.

LG AI Research expanded the supported languages for K-EXAONE 2.0 from Korean, English, Spanish, German, Japanese and Vietnamese to 10 by adding French, Italian, Portuguese and Polish.

An LG AI Research official said, “It supports the largest number of languages among domestic AI foundation models, laying the groundwork for sovereign AI to target not only Korea but also global markets.”

LG AI Research stressed it also worked to improve stability, alongside performance, by complying with Korean special characteristics and global ethical standards.

Im Woo-hyung (임우형), co-head of research at LG AI Research, said, “K-EXAONE 2.0 is not simply a model with a large number of parameters. It is meaningful in that domestic researchers independently completed the entire process, from designing a 750B-class model to data training, distributed training and building an inference environment.” He added, “We have secured the capability to compete in the same weight class as global frontier models, which is the core goal of the Independent AI Foundation Model Project.”