LG AI Research Institute K-EXAONE performance
LG has unveiled an AI model with performance surpassing that of the United States and China, which are leading countries in the artificial intelligence (AI) field. The move is seen as demonstrating technological capabilities that can compete with the latest AI models from global big tech companies.
LG AI Research Institute on the 30th disclosed the performance of its independently developed AI model “K-EXAONE (EXAONE)” at the “1st Presentation of Independent AI Foundation Models” hosted by the Ministry of Science and ICT at COEX in Gangnam, Seoul. The institute stated that it scaled up K-EXAONE into a frontier-level model with 236B (236 billion parameters). It also explained that it applied differentiated technologies to achieve both efficiency and performance.
Specifically, LG AI Research Institute emphasized that K-EXAONE applied new AI technologies to enhance inference efficiency compared with the existing EXAONE 4.0, while reducing memory requirements and computation volume. It stated that it secured both performance and efficiency through a proprietary Mixture-of-Experts (MoE) architecture and, by adding Hybrid Attention technology, reduced memory requirements and computation volume by 70%.
An official at LG AI Research Institute explained, “By designing the model to run in an A100-class GPU environment instead of using high-cost infrastructure, we significantly lowered the burden of build-out and operating costs, thereby enabling startups and small and mid-sized enterprises to easily adopt frontier-level AI models.”
LG K-EXAONE consortium structure
K-EXAONE overwhelms competing models in performance evaluation… “At global top-5 level”
LG AI Research Institute stated that it pursued the development of K-EXAONE with an ambitious goal of achieving more than 100% of the performance of the latest global AI models. At this presentation, it announced that among frontier-level open-weight models (with publicly available weights) of similar scale from the United States and China, it had set Alibaba’s “Qwen3 235B” as the first performance target model. In the first evaluation, K-EXAONE recorded an average score of 72.03 across 13 benchmark tests, achieving 104% of the performance evaluation score compared with the first target model Qwen3 235B (69.37 points). It also showed superior performance in comparison with the latest open-weight model from U.S. OpenAI, GPT-OSS 120B (69.79 points).
OpenAI’s GPT-OSS 120B model and Alibaba’s Qwen3 235B model are currently ranked 6th and 7th globally among open-weight models in the Intelligence Index assessment conducted by Artificial Analysis, a global AI analytics firm.
The K-EXAONE consortium at the 1st Presentation of Independent AI Foundation Models hosted by the Ministry of Science and ICT at COEX in Gangnam, Seoul, on the 30th
In the first evaluation, LG AI Research Institute’s K-EXAONE effectively secured performance that could place it among the global top five open-weight models. LG AI Research Institute emphasized that, based on its AI foundation model development capabilities accumulated over the past five years, it completed K-EXAONE in just five months. It plans to develop models that can fully compete with the highest-tier models from global big tech companies, which have parameter counts in the trillions.
An official at LG AI Research Institute said, “K-EXAONE has achieved the ambitious goal of delivering performance exceeding 100% of the latest global AI models,” adding, “We will continue to advance K-EXAONE’s performance based on LG’s differentiated technologies to contribute to strengthening national competitiveness.”
Koo Kwang-mo, Chairman of LG Group
Meanwhile, LG Group Chairman Koo Kwang-mo previously stressed innovation in his 2026 New Year’s address. Chairman Koo said, “At this inflection point where a new future is opening, we must move beyond the ways that have brought success so far and leap forward through new innovation,” adding, “For innovation, thinking and behavior must change, and choice and focus are the starting point.” He further emphasized that intense focus ultimately differentiates customer experience and completes the value that raises expectations to a new level.
In line with this, LG AI Research Institute plans to elevate its AI technology capabilities to a level others might consider impossible. It intends to concentrate its capabilities on creating and demonstrating LG’s unique innovative value that redefines the standard of the times. Ultimately, its goal is to secure competitiveness that surpasses global frontier AI models and become a game changer that leads the Republic of Korea to become one of the world’s top three AI powerhouses.
ⓒ dongA.com. All rights reserved. Reproduction, redistribution, or use for AI training prohibited.
Popular News