KT将于10月30日推出Token Factory，整合最多500个AI模型并宣称降低最多25%令牌消耗

KT (030200.KS) is launching Token Factory, a platform that will bring together up to 500 artificial intelligence models from around the world in a single place. The service lets customers handle sign-up, usage tracking and cost management in one place so they can pursue AI transformation tailored to their business goals, budgets and security needs. With token consumption soaring in the era of AI agents, KT says savings of up to 25% are the platform's main selling point.
KT unveiled its plans for KT Token Factory at a press briefing on the 27th at the Gwanghwamun West Building in Seoul's Jongno District. Token Factory is, in simple terms, a platform that allows users to tap multiple AI models without signing up for each one separately. "Competition in the AI market is rapidly shifting from who builds the better model to who makes it easier and more efficient for others to use AI," said Park Sang-won, head of KT's AX Business Group. "Token Factory extends the connectivity, billing and protection capabilities we built up in the telecom business into AI, and we will develop it into a core platform for Korea's AI ecosystem."
KT plans to launch Token Factory on the 30th of this month and expand its customer base and service scope in stages. At launch, it will support 73 models in total — seven that KT operates directly on domestic infrastructure and 66 global models. From next month, it will link more than 500 models through a tie-up with the global AI platform OpenRouter. Customers will be able to choose the technology that suits them, including Korean-built models, open-source models and frontier models. "The seven models are a selection of the most widely used models in Korea and the latest models from domestic startups," said Lee Jin-hyung, head of KT's AX Business Division. "Next year we aim to expand sovereign AI models to about 10."
KT has improved cost efficiency by applying Auto Model Router, its in-house AI model routing technology. The technology analyzes the task type and difficulty of a query along with the quality required and the cost involved, then automatically selects the most suitable AI model. The platform also applies prompt compression technology, which strips out unnecessary repetition while keeping the meaning of the query and the quality of the response largely intact. KT says prompt compression and model routing together can cut token consumption by up to 25%.
Protection of customer data is another selling point. Companies can configure the service to use only domestically operated models or to exclude specific models, depending on the nature of their work. "We also support guardrails that filter out personal information and harmful language, as well as company-specific banned terms and restrictions on access by IP address," said Won Man-ho, head of KT's AX Platform Division, adding that the service offers a range of features to keep client confidential information from being exposed.
At launch, pricing will be split into Standard and Enterprise plans. Standard is a pay-as-you-go arrangement billed according to usage. Under Enterprise, companies that commit to a certain volume of use receive lower rates and additional features. KT will calculate charges based on records of token use for each request, and customers will be able to review usage and costs by user and by model. The company also plans to offer combined billing of AI charges with corporate telecom bills in the future. Starting with corporate clients, KT plans to extend the service to individuals and developers in March next year and to introduce subscription plans.
KT intends to use Token Factory to accelerate AI transformation among clients in the financial and public sectors. For the financial sector, where security and data management requirements are stringent, it will supply models operated within Korea. In the public sector, it plans to deploy Token Factory in KT Cloud's public-sector cloud environment and to draw on its own AI foundation models as well as vetted open-weight models. "We see the main revenue source as the margin generated from producing tokens directly through KT data centers, rather than simple brokerage fees," Lee said.