▲ Park Sang-won, senior vice president and head of KT's AX Business Division, delivers opening remarks during a press conference on KT's AI Transformation (AX) business strategy held at the KT Gwanghwamun West Building in Jongno-gu, Seoul, on the 27th.
KT is set to launch "Token Factory," an AI platform that allows users to access both domestic and international artificial intelligence (AI) models in one place while integrally managing usage and costs, on the 30th.
In its initial launch phase, the platform will provide 73 AI models, with plans to integrate more than 500 models next month by partnering with global AI platform OpenRouter.
Starting with enterprise customers, KT plans to expand the service targets to individuals and developers by March next year and introduce subscription-based pricing plans.
KT held a press conference today (the 27th) at the Gwanghwamun West Building in Jongno-gu, Seoul, to unveil "KT Token Factory," which comprehensively provides everything from AI model selection and cost management to security functions.
Token Factory is a platform that allows users to select AI models suited for their work purposes and costs, and integrally manage everything from usage monitoring to payment.
KT has enhanced cost efficiency by applying its proprietary AI model routing technology, Auto Model Router.
This technology comprehensively analyzes the task type, difficulty, required quality, and cost of a query to automatically select the appropriate AI model.
According to KT, Auto Model Router ranked first overall in the global benchmark "RouterArena."
Prompt compression technology tailored to the characteristics of the Korean language was also applied.
This method removes unnecessary repetitive expressions to reduce the number of input tokens while maintaining the meaning of the query and the quality of the response at a similar level.
KT explained that prompt compression and model routing technology can reduce token usage by up to 25%.
It noted that prompt compression cuts usage by about 20%, while selecting an appropriate low-cost model cuts about 5%.
Among the 73 AI models provided at the initial launch, 7 are sovereign AI models for which KT directly provides inference services using graphics processing units (GPUs) at domestic data centers.
Lee Jin-hyung, senior vice president and head of KT's AX Business Headquarters, said, "These 7 models were selected from the most widely used models in South Korea and the latest models from domestic startups," adding, "Next year, we aim to expand the number of sovereign AI models to around 10."
In addition, KT announced that it has signed an official partnership, the first of its kind in South Korea, with OpenRouter, a global AI model marketplace operator.
In November, KT will link more than 500 global AI models offered by OpenRouter to Token Factory, while also pursuing a business to distribute domestic AI models overseas through OpenRouter.
However, depending on the service provider, inference for global models may take place on overseas servers.
KT plans to disclose information on the provider and operating region for each model to users.
(Photo: Yonhap News)
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.