KT Token Factory cuts rates 30% for seven locally hosted AI models
Edited and reviewed by AI Ping Editorial Team
3-line summary
- Seven KT-hosted models receive discounted token rates through November 30.
- The public catalog lists seven KT-hosted and 65 global entries with distinct prices and processing conditions.
- Prepaid service and additional workspaces are planned for December, with individual developer access in 2027.
KT Token Factory began discounting input and output token rates by 30% for seven domestically hosted AI models on October 30. The promotion excludes global models and ends November 30 at 11:59 p.m. KST.
The enterprise service connects multiple models through one API and consolidates usage and billing in Korean won. Its public homepage labels automatic model routing as a beta feature: KT says it considers the purpose of a request alongside quality, speed and cost when selecting a model.
The public catalog showed 72 entries when checked on October 30, split into seven KT-hosted and 65 globally hosted entries. Listings show prices, context capacity and provider information. KT says its counts include model versions and that availability can vary with contracts and provider policies.
The catalog listed discounted Mi:dm K 3.0 Pro rates at $0.14 per million input tokens and $0.56 per million output tokens. Solar-Pro-4 was listed at $0.21 and $0.84 respectively. Charges therefore depend on the chosen model and the mix of input and output, even at the same total token volume.
Promotional dollar rates exclude VAT. Bills use won rates recalculated daily at the previous business day’s exchange rate. Customers need corporate verification and a product contract.
Tokens are the units used to process prompts and generated answers. The FAQ says input and output are charged separately, with some models also pricing cached input. It lists no additional fee beyond model charges and VAT. Postpaid invoices and tax invoices are issued early in the following month; payment options include automatic card payment, bank debit and giro.
Domestic hosting and overseas processing
The service-opening notice dates the opening to October 29. It distinguishes KT-hosted domestic models from global providers, whose processing may take place overseas. Registered users receive a limited number of free model trials before contracting; API access follows corporate verification and a completed contract.
The privacy notice, effective October 29, states that prompts and input data for models directly hosted by KT are processed domestically. It specifies 30-day retention for information used to generate responses, including conversations, attachments and generated material. Token usage and settlement metadata have a separate five-year retention period. The overseas-transfer table lists the US for OpenRouter and OpenAI, and countries including the US, Singapore, Ireland and the Netherlands for Microsoft. It says external providers may retain information on overseas servers under their own policies.
The product page offers usage-based Standard service and an Enterprise option with negotiated commitments and support. Standard postpaid usage is settled monthly; Enterprise service levels are negotiated. The comparison table marks zero data retention and region restrictions as unavailable in Standard and planned for Enterprise.
The FAQ describes compatibility with the OpenAI API format, subject to each model’s supported features. Provider errors or timeouts can trigger another provider or a configured fallback model; billing is based on the final successful response. The dashboard can track costs by model and API key and set an organization’s budget.
Prepaid service and additional workspace management are scheduled for December 2026. The opening notice also places administrator controls over prompt and response storage among future additions. The homepage schedules service for individual developers in 2027.
