Nam Hyun-woo has worked as a staff writer at The Korea Times since 2013, mostly covering business and politics. He currently belongs to the Business Desk where he covers topics such as emerging tech, AI, ICT and Korea's chaebol community. Prior to joining the team, he was the paper's correspondent for the presidential office of Korea during the Yoon Suk Yeol and Moon Jae-in administrations.
LG unveils ecosystem of self-developed EXAONE AI
LG AI Research Co-President Lim Woo-hyung speaks during the LG AI Talk Concert event at LG Science Park in western Seoul, Tuesday. Courtesy of LG AI Research
By Nam Hyun-woo
LG Group’s artificial intelligence (AI) control tower, LG AI Research, unveiled on Tuesday the ecosystem of its self-developed EXAONE AI models, which encompasses the foundational EXAONE series and a slew of variants designed to perform agentic roles in various industries.
During an event at its headquarters in Seoul, LG AI Research presented a blueprint for EXAONE models, which have evolved based on AI capabilities that the institute has built up over the past five years.
“We aim to secure global competitiveness based on our own foundation model, and go beyond developing AI models by applying them across various industries to achieve both versatility and specialization,” LG AI Research Co-President Lim Woo-hyung said. “We will also build a broader AI ecosystem in collaboration with global partners.”
EXAONE is a large language model (LLM) that LG AI Research first unveiled in 2021. Since then, the model has gone through a series of upgrades and the institute earlier this month showcased EXAONE 4.0, a hybrid model that combines generative LLM and reasoning AI, and can form and test hypotheses independently to solve problems.
During the event, Co-President Lee Hong-lak introduced EXAONE 4.0 and EXAONE Path 2.0, a precision medical AI model specializing in histopathology image analysis that can reduce the time it takes to diagnose a disease from two weeks to under one minute.
Along with the two models, Lee for the first time showcased EXAONE 4.0 Vision Language (VL), a multimodal AI model capable of comprehending technical documents and images as complex as molecular structures.
LG AI Research Co-President Lee Hong-lak presents EXAONE 4.0 during the LG AI Talk Concert event at LG Science Park in western Seoul, Tuesday. Courtesy of LG AI Research
Dubbed “the eye of EXAONE,” Lee described the VL model as “a multimodal foundation model with exceptional image comprehension capabilities.”
“In the AI transformation that industries are striving for, the most crucial first step is the ability to accurately understand the vast volume of internal documents companies own,” Lee said. “This goes beyond simply processing text to understand various formats such as tables and charts to deliver insights users truly need. EXAONE 4.0 VL has achieved world-class performance in understanding charts and infographics embedded within documents.”
LG AI Research said that EXAONE 4.0 VL outperformed rival models such as Llama 4, InternVL3, and Qwen2.5-VL in understanding English-based charts, documents and images, in addition to Korean-language documents. It also demonstrated comparable performance to top-tier models in general tasks.
In particular, the model achieved a global top score of 92.8 in ChartQA, a benchmark that evaluates visual and language comprehension of charts, surpassing Llama 4 Maverick’s score of 90.
LG AI Research Co-President Lee Hong-lak presents EXAONE 4.0 Vision Language during the LG AI Talk Concert event at LG Science Park in western Seoul, Tuesday. Courtesy of LG AI Research
Along with the models, the institute also unveiled ChatEXAONE, an enterprise AI agent based on EXAONE, as well as EXAONE Data Foundry and EXAONE On-Premise.
EXAONE Data Foundry is an AI-powered data factory that generates high-quality datasets.
“Many companies want to develop industry-specific AI models, but the biggest challenge is securing sufficient and specialized training data,” said Choi Jung-kyu, head of LG’s AI Agent Group. “Data Foundry helps a single person generate a dataset in just one day — a task that would normally take 60 experts three months to complete.”
LG AI Research said that pilot projects conducted with LG affiliates and government-funded institutions showed that data productivity increased by at least 1,000 times, while data quality improved by an average of more than 20 percent compared to conventional methods.
As a turnkey solution for firms seeking AI transformation, the institute also introduced EXAONE On-Premise, which allows companies to build enterprise AI agents in an isolated environment, thus enabling them to use EXAONE models without security concerns.
The On-Premise solution will include access to EXAONE models, Data Foundry, various other tools and dedicated AI accelerators.
FuriosaAI CEO June Paik speaks during the LG AI Talk Concert event at LG Science Park in western Seoul, Tuesday. Courtesy of LG AI Research
June Paik, CEO of FuriosaAI, which partnered with LG for the On-Premise solution by providing its neural processing unit (NPU), said, “It is rare to find examples globally where a solution is built on NPUs.”
“LG was the first to begin testing our second-generation chip, Renegade, after its release in August last year, and the performance optimization target was achieved in June this year,” Paik said.
Looking ahead, Lee said, “The global AI industry is evolving toward agentic AI, where systems are capable of making decisions and taking actions on their own.”
“This evolution will eventually lead to physical AI, a technology that can change the real world. We will accelerate the development of physical AI, which enables systems to perceive, reason and interact with real-world environments.”