0
LG AI Research announced on the 7th that its foundation model "EXAONE" has achieved top-tier performance in the global market through various industry-specialized versions. The photo shows LG's headquarters in Yeouido. / LG

LG's AI foundation model "EXAONE" has outperformed models from major U.S. and Chinese tech companies in a global evaluation. Beyond a general-purpose language model, LG plans to expand EXAONE into industry-applicable versions while gradually broadening the types of data it can process, aiming to build competitiveness in the global AI market.

LG announced on the 7th that "EXAONE Tabular," a tabular data prediction model developed by LG AI Research, and "EXAONE Forecast," a future data prediction model, each achieved state-of-the-art (SOTA) performance on separate global leaderboards — surpassing models from Google in the U.S. and Alibaba in China.

These global leaderboards evaluate and verify the predictive capabilities of AI foundation models using real-world data from various industries, including chemistry, healthcare, finance, manufacturing, sales, and marketing. The achievement demonstrates that LG AI Research has established global competitiveness not just in general-purpose language models, but in specialized AI that can be directly applied on industrial sites.

Lim Woo-hyung, co-head of LG AI Research, said global tech giants' focus is rapidly shifting from general-purpose language models toward industry-specialized AI foundation models that solve real-world problems on the ground, adding that this achievement demonstrates how data accumulated across manufacturing and other industrial fields can become a distinctive competitive edge for Korean AI.

EXAONE Tabular ranked first overall in categorical data prediction on TabArena, a leading global leaderboard. Its evaluation score (ELO) of 1,760 surpassed Google's latest model, TabFM, released last month, which scored 1,749 — demonstrating LG's technical competitiveness in specialized industrial AI.

LG AI Research said EXAONE Tabular expands the model's scope beyond the text and image domains where global tech giants are competing for performance, into processing and predicting tabular data widely used in industrial settings. Since most data generated in industrial environments takes the form of tables organized in rows and columns, this marks a meaningful expansion of AI's applicability into structured data — an area traditional language models have struggled to handle.

LG AI Research plans to conduct proofs of concept (PoC) for EXAONE Tabular in manufacturing, bio/healthcare, and finance in the second half of this year, with plans to translate these into actual business applications. In manufacturing, for example, the model could analyze inspection data such as battery cell voltage and internal resistance to detect defects, while in the bio sector, it could predict risk levels and treatment response based on patient clinical data. In finance, it's expected the model could be used to predict abnormal transactions and the likelihood of delinquency or default based on data analysis.

		0
LG's TabArena leaderboard ranking. / LG

LG AI Research also achieved top-tier predictive accuracy on GIFT-Eval, a global leaderboard developed by Salesforce, the world's leading customer relationship management (CRM) company, used to evaluate time-series forecasting performance across various industries. EXAONE Forecast, LG's time-series foundation model, outperformed models from Google and Alibaba, demonstrating zero-shot performance — the ability to predict data from previously unseen domains without additional training — establishing its versatility.

LG AI Research said the results reflect experience accumulated from consistently applying time-series forecasting technology in real industrial settings. Working with affiliates such as LG Electronics and LG Energy Solution, the institute has developed AI models to forecast seasonal product demand and raw material prices, including lithium, and applied them to actual operations. Starting this year, it also expanded into financial forecasting, launching a stock market prediction AI service with Koscom and the London Stock Exchange Group (LSEG) that analyzes about 8,000 stocks listed in Korea and the U.S. daily, providing prediction scores and analytical commentary for each stock.

Through this approach, LG AI Research is expanding EXAONE's scope of application, centering on industry-specialized foundation models tailored to real-world industrial data and business needs. As it broadens the types of data it processes, the institute is also developing a robotics foundation model capable of understanding visual information.

Lee Soon-young, head of LG AI Research's Data Intelligence Lab, said LG's differentiated competitiveness lies in the data it has accumulated across various industries and its deep understanding of industrial settings, adding that LG will continue expanding its portfolio of industry-specialized foundation models to build a new axis of competition in the global AI market.

                                                                                                              Lee Ji-sun