LG Unveils Korea’s First Open-Source AI Model: EXAONE 3.0

LG AI Research has launched Korea’s first open-source artificial intelligence model, EXAONE 3.0, to advance the AI research ecosystem. This release includes a versatile and cost-efficient “lightweight model” designed for various research applications.

In addition, LG Group has introduced an open beta version of ChatEXAONE, an enterprise AI agent built on EXAONE 3.0, to its employees. This new tool is expected to enhance daily work tasks and support the company’s transition to the “era of everyday AI.”

EXAONE 3.0 demonstrates advancements in performance and efficiency compared to its predecessors, EXAONE 1.0 and EXAONE 2.0. The latest model requires 56% less processing time for inference, 35% less memory, and 72% lower operating costs.

These improvements are attributed to LG AI Research’s focus on lightweight and optimization technologies, which reduced the model’s size by 97% while enhancing its performance.

The technical report reveals that EXAONE 3.0 achieved top rankings in 13 benchmark categories, including real-world applications, coding, and mathematics. This performance places it on par with other leading open-source AI models, such as Meta’s Llama 3.1 and Google’s Gemma 2.

Notably, EXAONE 3.0 excelled in its bilingual capabilities, particularly in Korean, showcasing its advanced linguistic proficiency.

For EXAONE 3.0, LG AI Research utilized over 60 million data sets from diverse fields, including patents, software code, mathematics, and chemistry. The company plans to expand this dataset to over 100 million by the end of the year, incorporating additional fields like law, biology, and medicine to further enhance the model’s capabilities.

In addition to its performance, EXAONE 3.0’s open-source release is strategic for LG Group’s aim to lead in the AI sector. By offering the lightweight model for free, LG hopes to foster support adoption and integration across industries.

In the latter part of the year, LG AI Research intends to embed EXAONE 3.0 into a range of products and services throughout its affiliates. The model will be offered in various configurations, including an ultra-lightweight version for on-device use and a high-performance variant for specialized applications. This strategy is designed to foster innovation across LG Group’s various business sectors.

ChatEXAONE, currently in open beta for LG Group employees, is an AI tool offering functionalities such as real-time web-based Q&A, document and image-based Q&A, coding support, and database management. The service will be refined through regular updates based on user feedback, with a full rollout planned as each affiliate prepares for deployment.

ChatEXAONE leverages Retrieval-Augmented Generation (RAG) technology to provide answers based on real-time web search results, ensuring responses reflect the latest information. This capability improves work productivity by assisting with tasks like searching, summarizing, translating, and coding. It supports queries in 22 programming languages and SQL, enhancing its utility for developers and data analysts.

LG AI Research is also pursuing global partnerships to expand the reach and impact of EXAONE 3.0. By making its model available for academic and research use, the company aims to enhance the AI research ecosystem and boost national AI competitiveness.

According to: koreatechtoday