KT is set to unveil a platform that allows users to access multiple AI (artificial intelligence) models in one place and pay based on usage. The service will launch on the 30th with 73 models, expanding to more than 500 by November. The company aims to reduce AI usage costs by up to 25% through its proprietary technology.
On Tuesday, KT held a press conference at the Seoul Jongno-gu Gwanghwamun West Building to introduce its AI platform, "KT Token Factory." The service integrates multiple AI models into a single API (Application Programming Interface), providing end-to-end support from model selection and usage monitoring to cost management and billing.
Recently, companies have been leveraging different AI models depending on their tasks, such as translation, summarization, coding, and reasoning. However, the burden of registering separately for each model and managing costs is growing.
Park Sang-won, head of KT's AX Business Division (Senior Vice President), emphasized that it is crucial to enhance cost efficiency across AI infrastructure, models, and agents by "utilizing GPUs efficiently and selecting appropriate AI models to produce tokens at a lower price."
In its initial phase, the platform will offer access to a total of 73 models, including seven models directly provided by KT from domestic data centers. Models running domestically include KT's proprietary model as well as those from Upstage and Motif Technologies. By November, the service will expand to more than 500 models through integration with OpenRouter, a global AI model distribution platform.
The pricing plans come in two types: Standard and Enterprise. The Standard plan is a pay-as-you-go model based on token usage, supporting both prepaid credits and postpaid billing. The Enterprise plan offers discounts based on usage scale and contract duration, while also providing customized features such as dedicated lines and SLAs (Service Level Agreements).
Lee Jin-young, head of KT's AX Business Department, stated, "The Standard plan primarily offers a postpaid settlement method where users pay for what they use." He added, "We plan to expand the service in March next year so that individual developers and general users can also access Token Factory, and we will launch a subscription-based pricing plan at that time."
Cost reduction is achieved through KT's self-developed "Auto Model Router" and prompt compression technology. The system automatically selects the appropriate AI model based on the difficulty of queries and required quality, while reducing unnecessary input content. KT explained that internal testing confirmed a cost reduction effect of approximately 20% from prompt compression and about 5% from automatic switching to more affordable models.
Data security has also been strengthened. KT applies data non-storage and non-learning principles to the seven models it directly operates in domestic data centers. The company also provides its proprietary guardrail technology for detecting personal information, sensitive data, and harmful expressions.
KT is prioritizing revenue generation through its own infrastructure over AI model brokerage fees. Lee stated, "Global AI model brokers typically charge a fee of around 5%, but KT's primary revenue model is not based on fees." He revealed, "We plan to secure revenue by directly running AI models using the GPUs (Graphics Processing Units) in our own data centers."
KT plans to integrate Token Factory with existing enterprise services such as communication lines and private clouds. The company is currently preparing pre-deployment with some businesses, and customer names are expected to be disclosed in the future.
in TECH
