KT will launch its AI platform, "KT Token Factory", which lets users choose among various AI models while managing usage and costs in an integrated way. KT plans to foster it as a core platform supporting corporate AI transformation (AX) and as a new growth business, to help expand AI infrastructure in South Korea.

KT held a media briefing on Oct. 27 at the Gwanghwamun West building in Seoul's Jongno district and unveiled Token Factory, which integrates AI model selection, cost management and security features.

Park Sang-won (박상원), head of KT's AX business division, said, "Now, a key question is how well AI is used and how it creates real business value." He said the ability to connect complex tasks and data and use AI organically will determine business performance.

◆Starts with 73 models…expands to more than 500 by year-end

Token Factory is a platform that helps companies select and use various AI models depending on business purpose, cost and security requirements. Instead of subscribing to or managing multiple AI services separately, users can handle everything from selecting models to checking usage, managing costs and making payments on a single platform.

KT presented "execution, efficiency and trust" as the core values of the Token Factory business. It aims to make it easier to apply various AI models to work and optimise token usage and costs. At the same time, it plans to boost data security to secure trust.

Lee Jin-hyung (이진형), head of KT's AX business headquarters, said, "Token Factory is a platform that supports the effective use of AI by connecting various AI tools." He said it enables companies to check usage across multiple AI models in one place and pay in Korean won through a unified billing system.

Token Factory, set to launch on Oct. 30, will include a total of 73 models, including 7 provided through KT's domestic infrastructure and 66 global AI models.

Models provided through domestic infrastructure include KT's "Mideum K 3.0", Upstage's "Solar Pro 4" and Motif Technologies' "Motif 3". It will also offer open-weight models such as Google's Gemma, Nvidia's Nemotron, Moonshot AI's Kimi and Z.ai's GLM. Global models supported include OpenAI's GPT series, Anthropic's Claude, and models from Mistral AI, DeepSeek, xAI and Alibaba Cloud.

KT has signed a partnership with OpenRouter, a global AI model distribution platform, and is expanding its model supply chain. It plans to increase the number of AI models available on the platform to more than 500 by year-end.

◆Automatic AI model selection and prompt compression…cuts token costs by up to 25 percent

Token Factory's core strengths are automatic AI model selection and token usage optimisation technology. Companies currently use different AI models depending on the nature of their work. But using high-performance models for simple questions or entering unnecessarily long prompts can increase token usage and costs.

To address this, KT applied its in-house "Auto Model Router". The technology automatically selects an appropriate AI model by analysing question type and difficulty, required quality and cost. Users can set the desired balance between cost and quality instead of specifying an AI model directly. The system adjusts selection by choosing a high-performance model when quality is prioritised and a relatively cheaper model when cost savings are prioritised.

KT said Auto Model Router ranked first overall in RouterArena, a global AI model routing performance evaluation. It also offers a "fallback model" feature that automatically switches to another pre-designated model if a request fails to process.

KT said its technology can reduce AI token usage by up to 25 percent. Kim Joon-seok (김준석), head of the AI Agentic Lab at KT's Future Technology Institute, said compression technology that removes unnecessary content from prompts can reduce costs by about 20 percent. He said routing technology that selects low-cost models suited to tasks can cut an additional roughly 5 percent.

Security has also been improved by screening verified AI models and not storing customer data or using it for training. KT applies its own guardrail technology that detects personal data, sensitive information and harmful expressions during AI input, and verifies generated answers.

◆Applies telecom billing technology to AI…postpaid plans and corporate-only services

KT applied billing technology accumulated in its telecommunications business to Token Factory. It uses "Token Detail Records (TDR)", an expansion of the "Call Detail Records (CDR)" concept used to record usage in telecom services, adapted for an AI environment. It precisely records the AI models used by customers and their token usage, and bills costs in an integrated way.

Pricing will be launched first with Standard and Enterprise plans. Standard will operate on a pay-as-you-go basis, charging for tokens actually used, and supports postpaid settlement. Enterprise for corporate customers will operate as a postpaid contract-based plan. It also provides additional features such as dedicated networks, private environments and service-level agreements (SLA) tailored to each company's requirements.

KT will start the service focused on business-to-business (B2B) transactions. By the end of December, it will expand the model lineup and add-on management functions for corporate use. In March next year, it plans to broaden the service to individual developers and general users. It will also review introducing subscription-based plans in line with entry into the individual user market.

KT also plans to diversify its revenue model through AI model brokerage fees and add-on services for corporate use. It aims to secure price competitiveness through discussions with AI model suppliers and generate revenue by combining add-on features such as private services.

Park said, "Based on Token Factory, we will support customers' AX transformation and develop it into a core platform that leads the growth of South Korea's AI ecosystem."