"K-EXAONE, which is participating in the government's national AI foundation model project, is delivering performance on par with OpenAI. We aim for the world's top tier." (Im U-hyeong, co-head of LG AI Research)
"Even for a Humanoid Robot ultimately built on the AI agent ixi-O, the method of communication is voice. We are preparing to apply various voice technologies to physical AI going forward." (Lee Sang-yeop, chief technology development officer at LG Uplus)
LG Uplus Chief Technology Development Officer (CTO) Lee Sang-yeop and LG AI Research co-head Im U-hyeong held a briefing on the 1st, a day before the opening of MWC 26, the world's largest mobile communications expo in Barcelona, Spain (local time), and stated accordingly the blueprint for "No. 1 AI One Team LG." EXAONE is built into LG Uplus' ixi-O.
K-EXAONE, developed in-house by LG AI Research, recently ranked first in phase one of the government's national AI project evaluation. However, it is not yet No. 1 on the global stage. According to Artificial Analysis, a global AI model performance analytics organization, in Jan. last year, based on open-weight models that disclose weights, K-EXAONE scored 32 points, placing seventh in the world. OpenAI was sixth with 33 points. LG plans to develop the highest-performing language model among existing global open-weight models during the second phase of its independent AI foundation model project in the first half of this year.
On this day, LG Group unveiled for the first time the future development roadmap for K-EXAONE (EXAONE), which is being developed as a national AI model. LG plans to push EXAONE development centered on four strategies: securing leadership in AI foundation model development; aiming for expert AI; expanding practical, on-site industrial applications; and ensuring trust and safety for sustainable AI.
LG AI Research co-head Im U-hyeong said, "The AI that LG seeks is not a race for higher intelligence but to build a partner that helps human life and solves problems in the real world," adding, "It is important to build good AI, but what we build is ultimately the key, and we will focus on creating AI that generates real value in the physical, real-world space beyond the stage of AX (AI transformation)."
Plans to link K-EXAONE with infrastructure going forward were also laid out. LG Uplus CTO Lee Sang-yeop said, "To implement AI that can help in the real world, powerful infrastructure and connectivity technologies to support it are essential," adding, "As K-EXAONE moves into the completion stage next year, One Team LG, including LG Uplus, is preparing Korea's largest AI data center (DC)." He went on to introduce the Paju AI DC, the largest in the greater Seoul area (200 MW), slated for completion in 2027, saying, "We are proceeding without a hitch to deliver the best experience with the world's top-performing AI."
The Paju AI DC can accommodate up to 120,000 GPUs. It is expected to become the execution site of the One LG strategy, which brings together the core capabilities of LG affiliates such as LG Electronics, LG Energy Solution and LG CNS. LG Uplus plans to deliver complete end-to-end, customized services based on K-EXAONE and the Paju AIDC.
In addition, LG plans to showcase infrastructure that combines LG AI Research's EXAONE, LG Uplus' enterprise AI platform and FuriosaAI's Neural Processing Unit (NPU) to demonstrate the synergy between K-AI and K-semiconductors.
LG AI Research plans to soon release EXAONE 4.5, a Vision-Language Model (VLM), as an open-weight model as it nears completion. A vision-language model combines language and visual intelligence to communicate by understanding and analyzing text and visual information in a multimodal manner, like a human. The company said it expects EXAONE 4.5 to become the highest-performing model globally among open-weight models of the same size while increasing the model's utility through modality expansion. In particular, starting with EXAONE 4.5, LG expects that the increasingly advanced EXAONE vision-language model technology will serve as the brain of KAPEX, a Korean-style humanoid, and establish itself as a core technology opening the era of physical AI.
On this day, LG Uplus also introduced a next-generation agentic AI strategy to be developed in cooperation with LG AI Research. LG Uplus CTO Lee Sang-yeop said, "As a telecom operator, we aim to evolve beyond merely connecting networks to becoming an AI software helper."
In addition, LG Uplus will build a Hybrid AI Infra architecture optimized to run agentic AI in real time. This architecture is expected to become a core technology that enables stable and cost-effective operation with minimal latency for large language models like EXAONE within corporate environments.
LG Uplus CTO Lee Sang-yeop said, "True competitiveness in the AI era does not lie solely in a volume offensive of bigger models and more compute resources," adding, "The deciding factor will be how precisely we design an agentic architecture that evolves on its own through a cyclical loop in which planning and execution, evaluation and refinement are repeated beyond single-pass inference."