KT said on Tuesday its self-developed large language model, Mideum K 2.5 Pro, received the AI Trustworthiness Certification 2.0 (CAT) from the Telecommunications Technology Association (TTA).

CAT is a private voluntary certification that evaluates governance reliability across the adoption and operation of AI. KT also received certification for the Pro version, following its earlier Mideum K 2.0 Base. KT explained it is the first case in South Korea in which an AI model in the same family has continuously passed reliability verification.

The newly certified Mideum K 2.5 Pro is a large language model with 32 billion parameters. It is a private, enterprise-focused model designed on the premise of use for business-to-business services, including work automation, customer response and internal tasks.

KT stressed that the certification is meaningful because it accumulates cases of reliability certification for domestic AI models and secures a system of continuous improvement and verification rather than a one-off certification.

"Mideum K 2.5 Pro" also proved competitive in performance compared with similar models. In the AAII (Artificial Analysis Intelligence Index) V3.0 benchmark, it posted top-tier performance against models in the same class such as Alibaba's Qwen (Qwen-3-30B-A3B) and South Korea's HyperCLOVAX (HyperCLOVAX-32B).

In the Tau-Square benchmark evaluation, which checks how well AI performs tasks by accessing real systems, it achieved results of about 87 percent.

Jae-hyung Park (박재형), head of KT's AX Future Technology Institute, said, "KT will continue to develop reliable AI models that customers can use with confidence and will take the lead in creating a responsible AI ecosystem that supports corporate customers' AI transformation."