KT Launches Token Factory With 500 AI Models

KT said on the 27th that it will launch KT Token Factory on October 30. The enterprise platform lets customers select, manage and pay for more than 500 AI models from Korea and abroad in one place. KT plans to develop the service as a growth business supporting customers’ AI transformation. Commercialization comes about four months after CEO Yoon-young Park proposed it as a new business in July, after taking office.

KT’s core cost-efficiency feature is its in-house automatic model router, which selects a suitable model based on a query’s task type, difficulty, quality and cost. KT said Korean-language prompt compression, designed to account for the characteristics of Korean, can reduce token use by up to 25%.

KT will provide AI inference directly and use Korean-optimized guardrails—tools that check inputs and responses for risks—to screen prompts for personal information and harmful language, as well as check AI answers. To build the service, KT formed an official partnership with global AI model marketplace OpenRouter. It also plans to expand its partner ecosystem through work with Uracle, CAST AI, Timely and Upstage.

Sang-won Park, executive vice president and head of KT’s AX Business Division, said competition in the AI market is shifting from building better models toward helping people use AI more easily and efficiently. KT Token Factory offers more than 500 integrated models, with automatic model selection and token-saving features.

KT said on the 27th that it will launch KT Token Factory on October 30. The enterprise platform lets customers select, manage and pay for more than 500 AI models from Korea and abroad in one place. KT plans to develop the service as a growth business supporting customers’ AI transformation. Commercialization comes about four months after CEO Yoon-young Park proposed it as a new business in July, after taking office.

KT’s core cost-efficiency feature is its in-house automatic model router, which selects a suitable model based on a query’s task type, difficulty, quality and cost. KT said Korean-language prompt compression, designed to account for the characteristics of Korean, can reduce token use by up to 25%.

KT will provide AI inference directly and use Korean-optimized guardrails—tools that check inputs and responses for risks—to screen prompts for personal information and harmful language, as well as check AI answers. To build the service, KT formed an official partnership with global AI model marketplace OpenRouter. It also plans to expand its partner ecosystem through work with Uracle, CAST AI, Timely and Upstage.

Sang-won Park, executive vice president and head of KT’s AX Business Division, said competition in the AI market is shifting from building better models toward helping people use AI more easily and efficiently. KT Token Factory offers more than 500 integrated models, with automatic model selection and token-saving features.