LG Unveils South Korea's First Open-Source AI Model

In a significant development, LG Group’s artificial intelligence focused research division, LG AI Research, unveiled the third generation of its hyperscale AI multimodal model Exaone 3.0 on Wednesday. This release marks South Korea’s first open-source AI model, aiming to enhance the AI ecosystem through improved performance and cost-efficiency.

Exaone 3.0 promises substantial advancements over its predecessor by requiring 56% less processing time, 35% less memory usage, and reducing operating costs by 72%. These improvements are expected to benefit a wide array of applications from academia to industry. LG AI Research has particularly highlighted its success in managing power consumption by reducing the model size by 97% while still enhancing performance compared to Exaone 1.0.

A notable feature of Exaone 3.0 is its capability as a bilingual model proficient in both Korean and English. It has outperformed global AI systems such as Meta’s Llama 3.1 and Google’s Gemma 2 in 13 benchmark scores, demonstrating superior performance in real-world use cases, coding, and mathematics. The model has been trained on over 60 million data points across diverse fields including patents, coding, math, and chemistry, and is on track to exceed 100 million data points by the end of the year.

The development of Exaone 3.0 reflects LG's strategic focus on AI as a future growth engine. The conglomerate has committed to investing 3.6 trillion won ($2.84 billion) in AI research and development by 2026. LG Group Chairman Koo Kwang-mo has emphasized the importance of bold investment and innovation in accelerating AI transformation across the company’s various business units.

LG AI Research plans to collaborate closely with its affiliates to integrate Exaone 3.0 into products and services by the second half of the year. Additionally, the research unit is exploring global partnerships to extend the reach and impact of the new AI model. Various model sizes tailored for different applications have been developed, ranging from ultra-lightweight for on-device AI to high-performance for specialized applications.

To facilitate a broader impact, LG AI Research intends to make the lightweight version of Exaone 3.0 available as an open-source resource. This initiative is aimed at fostering AI research and innovation by making advanced generative AI technologies accessible to academia and research institutions. President Bae Kyung-hoon of LG AI Research stressed the importance of creating AI solutions that are practically applicable in real-world industries.

In tandem with the Exaone 3.0 launch, LG Group has started an open beta service of ChatExaone, an enterprise AI agent, for its affiliates’ employees. This AI agent can perform a range of tasks such as real-time web-based queries, document and image-based questions, coding, and database management, significantly enhancing work productivity. Plans are in place to extend this service and develop a mobile application based on user feedback.

LG’s commitment to AI is evident in the establishment of LG AI Research in December 2020, and the company has since been accelerating its AI capabilities. The strategic emphasis on AI has resulted in tangible benefits such as improving production processes, product development, and customer service. Through ongoing investments and innovations, LG aims to maintain a competitive edge in AI technology.