KT is launching Token Factory, a platform that gives businesses one place to access AI models, manage usage and pay by token. Its more than 73 models, automatic model selection and prompt compression make this a practical bid to simplify enterprise AI buying, not just another model launch.
Watch Desk analysis
What happened
KT plans to launch the service on Friday, initially offering 73 models through a single API. The Korea Times reports that the company plans to expand the range to more than 500 in November through a partnership with OpenRouter. KT says its router selects models according to task requirements and balances response quality against cost.
The service will initially target corporate and public-sector customers. Standard customers can use postpaid billing or prepaid credits; enterprise customers can negotiate lower rates in return for committing to usage volumes. The report says individual developers and consumers are planned for March 2027, when KT expects to introduce subscriptions.
Our top picks
- One API for multiple models
Businesses can use the platform to connect to models without separately signing up for and integrating each one. - Automatic model routing
KT says its system chooses a model for each task, aiming to balance quality and cost. - Prompt compression
KT says routing and compression can reduce token costs by up to 25 per cent, a company estimate rather than a reported independent result. - Flexible billing
Customers can choose prepaid credits or usage-based postpaid billing, while enterprise buyers can negotiate volume-based rates. - Domestic inference options
Seven launch models, including Solar, Mi:dm and Gemma, will run on infrastructure in South Korea.
Why it matters
The offer tackles a real enterprise nuisance: stitching together model access, usage tracking and billing across different providers. A single service could make it easier for organisations to try several models and keep an eye on consumption. KT also plans to connect Token Factory with its telecom, cloud and enterprise AI services, giving it a route to sell inference alongside infrastructure it already operates.
The cost claim deserves a proper test. Savings of up to 25 per cent would matter to teams using AI at scale, but actual results will depend on the task, chosen model and prompt. The headline percentage is a promise to measure, not a receipt.
Our read
This is a substantial enterprise platform launch, with model choice and billing controls that are more useful than a bare catalogue. KT’s strongest case will be whether its routing delivers competitive results at lower total cost, not how quickly the model count grows. A bigger menu is only helpful if the waiter knows what to recommend.
What to watch
- Whether the promised expansion to more than 500 models arrives in November.
- How Token Factory performs on real enterprise workloads, including quality and cost.
- Which models and workloads qualify for domestic inference.
- Whether individual access and subscription pricing arrive as planned in March 2027.
Discussion spark: For enterprise AI, is a single platform that chooses models and consolidates billing more valuable than direct access to each provider, or does that convenience risk obscuring cost and performance?
Sources and evidence
- Source update (27 October 2026, 06:53 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.