KT to Unveil Proprietary AI Model 'Mi:dm K 2.5 Pro' at MWC26

KT (030200) announced on the 26th that it will showcase the development journey and technical achievements of its proprietary artificial intelligence model "Mi:dm K" at the Mobile World Congress 2026 (MWC26), the world's largest mobile communications exhibition opening on March 2.

The model KT is presenting is "Mi:dm K 2.5 Pro." The model has been expanded to 32 billion parameters, enhancing knowledge density and reasoning performance. It supports input lengths of approximately 128,000 tokens, enabling analysis of documents spanning hundreds of pages. The system has also expanded from Korean and English to a four-language framework including Japanese and Chinese, increasing global applicability.

Mi:dm K 2.5 Pro recorded the highest performance among Korean AI models in the AAII v3.0 evaluation based on the platform's metrics. It scored 87% on the "Tau-Squared Bench," which measures agentic AI performance, demonstrating global-level tool utilization and task execution capabilities.

Through the "K Data Alliance," KT collaborated with public institutions, academia, media, and educational sectors to secure high-quality data reflecting Korean linguistic context. The data was processed through the company's proprietary refinement, processing, and verification pipeline for training. The model achieved consistently strong scores on benchmarks measuring complex reasoning and Korean language comprehension while reducing large-scale GPU infrastructure requirements, establishing competitiveness as an efficient and practical enterprise AI model.

Oh Seung-pil, Executive Vice President and Head of Technology Innovation at KT, said, "Mi:dm K is the core foundational technology of KT AI that has been validated in actual service environments, going beyond a research model." He added, "We will use MWC26 as an opportunity to demonstrate the competitiveness of Korean AI to the global market and establish ourselves as an agentic AI partner that provides practical support for enterprise customers' AI transformation (AX)."