LG AI Research Unveils EXAONE 4.0
Passes National Professional Exams Including Medical Licensing
Demonstrates World-Class Performance
LG AI researchers have unveiled 'EXAONE 4.0,' which has even passed the medical licensing exam, positioning themselves at the forefront of 'AI industrialization.' This is the first hybrid AI model developed in Korea, combining a large language model (LLM) with strengths in knowledge-based rapid response and a reasoning AI model equipped with problem-solving abilities.
LG AI researchers unveiled 'EXAONE 4.0,' which even passed the 'medical licensing exam,' leading the 'AI industrialization.' Getty Images
The expert model has demonstrated performance high enough to pass nationally certified professional exams, while the on-device model, which can be embedded in home appliances and robots, has surpassed the capabilities of OpenAI's GPT-4o mini. LG's strategy is to accelerate the adoption of AI technology in industry through the commercialization of EXAONE 4.0.
On July 15, LG AI Research announced the launch of EXAONE 4.0, Korea's first hybrid AI model. As of now, only companies such as Anthropic in the United States and Alibaba in China have released hybrid AI models. Even OpenAI, widely known for ChatGPT, is still in the process of developing its GPT-5 model as a hybrid model.
Baek Kyun LG AI Research Director is introducing 'ExaOne Deep' at Nvidia's annual developer conference (GTC) held in San Jose, California, USA, in March this year. Photo by LG Yonhap News
In benchmark tests, EXAONE 4.0 outperformed leading open-weight models from the United States, China, and France, demonstrating world-class performance. By evaluation category, it scored 92.3 points in general and reasoning abilities, 81.8 points in advanced professional knowledge, 66.7 points in coding skills, 75.4 points in scientific problem-solving, and 85.3 points in mathematical problem-solving.
EXAONE 4.0 is expected to deliver high-quality answers even in fields that require specialized knowledge. The model is divided into a 32B expert model (with 32 billion parameters) and a 1.2B on-device model (with 1.2 billion parameters). The expert model proved its expertise by passing the written exams for six nationally certified professional licenses: physician, dentist, Korean medicine pharmacist, customs broker, appraiser, and loss adjuster.
The on-device model, despite being half the size of the previously released EXAONE 3.5 (2.4B model) in December 2024, delivered superior performance in specialized evaluation metrics such as mathematics, coding, and science compared to OpenAI's GPT-4o mini. According to LG, it is among the best in the world for AI models of similar scale. Since the on-device model can be embedded in home appliances and robots, it is expected to enable LG to take the lead in these markets.
To usher in the era of AI industrialization, LG AI Research is moving forward with the commercialization of EXAONE 4.0. The company has begun providing API services in partnership with FriendlyAI. This allows everyone from individual developers to enterprises to utilize EXAONE without the need for high-performance GPUs. LG AI Research believes that releasing the API will serve as a turning point, accelerating the adoption of AI technology across various industries.
Lee Jinsik, Head of EXAONE Lab at LG AI Research, stated, "We will continue our research and development so that EXAONE can become Korea's representative frontier model," and emphasized, "We will prove our competitiveness in the global market."
Meanwhile, LG AI Research plans to hold the 'LG AI Talk Concert 2025' at LG Science Park in Seoul on July 22, where it will present its R&D achievements and future plans for AI technologies, including EXAONE 4.0.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
