LG AI Research Institute Unveils "K-ExaOne," Surpassing US and China Open-Source AI Models in Performance

LG AI Research Institute has unveiled its independently developed large-scale artificial intelligence model, "K-ExaOne," stating that it has demonstrated higher performance than the latest open-source AI models from the United States and China.

On December 30, at a government-hosted event held at COEX in Gangnam, Seoul, LG AI Research Institute revealed the performance of its proprietary AI model, "K-ExaOne."

This model was developed on a much larger scale than the previous ExaOne, with faster response generation speeds and reduced memory and computational requirements. The company explained that it has enhanced performance while lowering operational costs.

K-Exa One Performance. LG

LG AI Research Institute stated that, based on its accumulated development experience, it has elevated the model to a level capable of competing with the latest global big tech AI models within just five months. Moving forward, the institute plans to develop even larger, top-tier AI models to engage in full-scale competition.

Performance comparison results were also released. LG AI Research Institute measured performance based on models of similar size among the large-scale open-source AI models released in the United States and China, confirming that K-ExaOne surpassed them.

For its performance benchmark, LG AI Research Institute set Alibaba's "Qwen3 235B"-one of the open-source AI models from the United States and China-as its primary target.

In the comparison, "K-ExaOne" scored an average of 72.03 points across 13 evaluation categories, outperforming Qwen3 235B. It also showed superior performance compared to OpenAI's open-source model, GPT-OSS 120B.

Both Qwen3 235B and GPT-OSS 120B are classified among the top open-source AI models in global evaluations. LG AI Research Institute explained that, based on these results, "K-ExaOne" is now close to being ranked among the world's top five open-source AI models.

K-ExaOne Consortium. LG

LG AI Research Institute explained that, leveraging five years of accumulated foundation model development expertise, it achieved performance capable of competing with the latest global big tech models within just five months of development. The institute plans to develop next-generation models with trillions of parameters to directly compete with the world's leading models.

A representative from LG AI Research Institute said, "We have achieved our goal of exceeding 100% of the performance of the latest global AI models," adding, "We will continue to enhance our capabilities based on differentiated technology and contribute to strengthening national competitiveness."

In his recent 2026 New Year's address, LG CEO Koo Kwang-mo emphasized, "At inflection points, we must leap forward with new innovations beyond existing methods of success," and, "Through selection and concentration, we must delve into levels that others consider impossible."

LG AI Research Institute plans to focus its research and development capabilities on securing competitiveness that surpasses global frontier AI models, aiming to position South Korea as one of the top three AI powerhouses in the world.