KT Launches Token Factory With 73 AI Models, Plans 500+ Expansion in November

KT unveiled its unified AI platform Token Factory on October 27, 2026, and launched enterprise service on October 30 with 73 initial AI models. Seven of those receive direct inference service on GPUs in KT's domestic data centers. KT plans to connect more than 500 models through OpenRouter in November; that figure is a future integration target, not the initial model count. Token Factory lets businesses manage model selection, usage monitoring, costs, and payments through one platform.

Image source: https://ax.kt.com/ax/solution/token-factory

Token Factory provides access to domestic and international AI models through a unified platform. Businesses can manage model selection, usage monitoring, costs, and payments without separate accounts or operating environments for each model. The platform includes KT's proprietary Auto Model Router and data guardrails.

Auto Model Router and Korean-Optimized Prompt Compression

Token Factory includes KT's proprietary 'Auto Model Router', which automatically selects an AI model based on user queries, task complexity, required quality, and costs.

The router analyzes user queries, task complexity, required quality, and costs. It assigns lightweight models to simple tasks and high-performance models to complex analysis.

• Korean Language Prompt Compression: KT says its Korean-optimized prompt compression removes unnecessary repetitive expressions from input.
• Combined Savings Claim: KT reported approximately 20% savings from Korean-language prompt compression and approximately 5% from selecting a suitable lower-cost model. Its maximum 25% figure refers to the combined approach, not prompt compression alone; it is a vendor-reported estimate.

KT described combining prompt compression and model selection to reduce token-related costs.

Data Protection Guardrails and KT-Hosted Domestic Inference

KT implemented guardrail technology in the platform.

According to KT, its guardrails inspect input prompts and model responses for personal information, sensitive content, and harmful expressions. These are among the data protection controls described for business users.

Seven of the initial 73 models receive inference on KT-operated GPUs in domestic data centers. That does not establish domestic processing for every global partner model: customers should check the specific model provider's data-processing region and terms.

OpenRouter Global Partnership and Rollout Timeline

KT also announced an official partnership with OpenRouter, a global AI model distribution platform.

KT expects to connect more than 500 OpenRouter models during November 2026 and to explore overseas distribution for Korean-developed models. These are announced expansion plans, separate from the initial 73 available models.

KT launched enterprise service on October 30, 2026. KT also said it aims to open the service to individual developers and consumers and introduce subscription plans in March 2027.

Sources

• The Elec: KT Unveils Token Factory Platform With Over 500 AI Models, Set for Oct. 30 Launch
• KT official product overview and architecture: KT Token Factory
• Yonhap, October 27: 73 models at launch; 500+ planned for November; savings breakdown
• Yonhap, October 30: Initial model count and expansion plan