LG AI Research released K-EXAONE 2.0 on Hugging Face on July 31, a 750-billion-parameter foundation model that triples the size of its predecessor and outperforms leading Chinese models on long-text comprehension benchmarks. It is the largest AI foundation model released in Korea, according to Seoul Economic Daily.

The model is part of Korea’s “Homegrown AI Foundation Model Project” overseen by the Ministry of Science and ICT. LG applied the Apache 2.0 open-source license, allowing free commercial use and modification.

Performance

K-EXAONE 2.0 scored 94.4 on the OpenAI-MRCR long-text comprehension metric and 89.6 on the Korean-language long-text benchmark Ko-LongBench. Both figures significantly exceed the 71.5 and 83.6 recorded by GLM-5.1, a major model from China’s Zhipu AI, according to Seoul Economic Daily.

“K-EXAONE 2.0 is not simply a model with a large number of parameters, but the result of domestic researchers independently completing the entire process from design to data training, distributed learning, and building the inference environment,” said Lim Woo-hyung, co-director of LG AI Research, according to Seoul Economic Daily. “We have secured the capability to compete in the same weight class as global frontier models.”

Open Source Strategy

The Apache 2.0 license means any company can use K-EXAONE 2.0 for commercial AI development without usage fees. LG AI Research plans to unveil an industry-specific AI foundation model next week, broadening its portfolio beyond general-purpose language and vision models into specialized industry knowledge, according to Seoul Economic Daily.

LG has also completed development of a public evaluation site for second-round review of homegrown AI foundation models, and plans to launch a separate service allowing public access to K-EXAONE 2.0.

Korea Enters the Foundation Model Race

K-EXAONE 2.0 is the second major open-source foundation model release from Asia in the same week, alongside Huawei’s openPangu-2.0-Pro (505 billion parameters). The simultaneous releases signal that the foundation model landscape is diversifying beyond the U.S. and China duopoly. For teams building agent systems on top of open-weight models, more competitive options from Korea, Japan, and Europe reduce dependency on any single provider’s model family and pricing decisions.