KT launches Token Factory to automate AI model selection and billing

A unified hub for AI models

KT is launching a new platform called Token Factory to help companies navigate the complex landscape of artificial intelligence. Instead of businesses having to subscribe to and manage multiple different AI services separately, this platform provides a single point of access. It integrates model selection, cost management, and security features into one interface.

The platform will initially offer 73 different models, which include a mix of domestic and global options. Local offerings include Mideum K 3.0 from KT, Solar Pro 4 by Upstage, and Motif 3 from Motif Technologies. Global options include well-known models from OpenAI, Anthropic, Mistral AI, and Alibaba Cloud.

KT plans to scale this library significantly by the end of the year, aiming to offer more than 500 models. To achieve this, the company has partnered with OpenRouter, a global distributor of AI models. This expansion is intended to help foster the AI infrastructure and business transformation within South Korea.

Automated routing and efficiency

One of the core features of Token Factory is its ability to automatically select the most appropriate AI model for a specific task. Many companies currently waste money by using high-performance models for simple questions or by inputting unnecessarily long prompts. KT's 'Auto Model Router' solves this by analyzing the difficulty and type of a user's request.

Users can set their preferences for the balance between cost and quality rather than picking a specific model manually. If a user prioritizes high quality, the system chooses a powerful model; if they prioritize savings, it selects a cheaper one. KT noted that this routing technology ranked first overall in the RouterArena, a global evaluation of model routing performance.

The platform also includes compression technology to remove unnecessary content from prompts, which can reduce costs by about 20 percent. Additionally, the routing technology itself can cut costs by roughly 5 percent. Together, these optimizations can reduce total token usage by up to 25 percent while providing a 'fallback model' if a primary request fails.

Telecom-style billing for AI

KT is applying its experience in telecommunications to simplify how businesses pay for AI usage. The company developed 'Token Detail Records' (TDR), which is an adaptation of the 'Call Detail Records' (CDR) used to track phone usage. This system precisely records which models were used and how many tokens were consumed, allowing for integrated billing.

The service will launch with two main tiers: Standard and Enterprise. The Standard plan operates on a pay-as-you-go basis where customers are charged for the tokens they actually use. This plan also supports postpaid settlement, which is often more convenient for corporate accounting than pre-paying for credits.

The Enterprise plan is designed for larger corporate customers who require more specific features. It offers postpaid contract-based plans along with dedicated networks, private environments, and service-level agreements. These additions are meant to meet the strict security and reliability requirements of large-scale business operations.

Security and data privacy

Security is a major focus for the Token Factory, as companies are often hesitant to share sensitive data with external AI models. KT plans to boost trust by screening only verified AI models for the platform. Furthermore, the company has stated that it does not store customer data or use it to train its models.

The platform includes its own guardrail technology to protect users during the interaction process. This technology is designed to detect personal data, sensitive information, and harmful expressions during the input phase. It also verifies the answers generated by the AI to ensure they are safe and appropriate for business use.

By combining these security measures with a unified platform, KT aims to make it easier for companies to adopt AI without compromising their data integrity. The goal is to provide a secure environment where businesses can experiment with various models while maintaining strict oversight of their information.

Roadmap for growth

KT will initially focus its Token Factory service on business-to-business (B2B) transactions. This allows the company to refine the platform's features and model lineup for corporate needs first. By the end of December, they plan to add more models and management functions specifically for these corporate users.

The company intends to broaden the service to include individual developers and general users starting in March of next year. At that stage, KT may also review the introduction of subscription-based plans to better suit the needs of individual consumers. This phased rollout allows the company to build a stable foundation before entering the broader consumer market.

Looking ahead, KT plans to diversify its revenue by collecting brokerage fees from AI model suppliers. They also intend to generate revenue through add-on services, such as private environments for corporations. Park Sang-won, head of KT's AX business division, stated that the platform will support customers' AX transformation.