KT's Token Factory goes live: 73 AI models behind one API, 30% off the seven it serves itself

South Korea's KT has gone live with its answer to a problem every company buying AI now faces: too many models, too many bills, no easy way to pick. Token Factory, the telecom operator's AI transformation platform, launched for corporate customers on October 30, putting 73 AI models behind a single API with unified usage tracking and billing in Korean won.

The launch comes with a sweetener. KT said on October 30 it will cut usage fees by 30% through November 30 on the seven models it serves directly from its own domestic infrastructure, for corporate clients that complete verification and sign a product contract. The aim, KT said, is to let companies try models hands-on and adopt the one that fits their operations at lower cost — lowering the barrier to corporate AI adoption.

What Token Factory does#

Token Factory covers the whole journey from model selection to payment: companies choose a model, use it, track usage, and settle the bill in one place, in won, drawing on KT's telecom billing machinery. It was unveiled at a media briefing on October 27 at KT's Gwanghwamun West Building in Seoul, where executives framed the shift in the AI market as moving from who builds the best model to who makes models easiest and cheapest to actually use.

The platform's centerpiece is its Auto Model Router, which picks the model for each task based on question difficulty, required quality, and cost, combined with prompt-compression technology that strips unnecessary content from prompts. KT says the combination can reduce token costs by up to 25% — a company-supplied claim, not an independently verified benchmark.

The 73 models at launch#

The launch catalog splits into two supply routes. Seven models run on KT's domestic infrastructure: KT's own Mi:dm K 3.0 Pro, Upstage's Solar-Pro 4, Motif Technologies' Motif 3, NVIDIA's NeMoTron 3 Ultra 550B A55B, GLM 5.3 Flash, Kimi K3, and Gemma 4 31B. Those seven are the ones covered by the 30% launch discount. The other 66 are global models, including OpenAI's GPT series, Anthropic's Claude, and models from Mistral, DeepSeek, xAI, and Alibaba Cloud.

KT plans to grow the catalog to more than 500 models by the end of 2026 through a partnership with OpenRouter, the global model-distribution platform — the first such agreement in Korea, KT says. KT also plans to earn revenue by running models on graphics processors in its own data centers, and says the platform will not store customer data or use it for training.

Why a telecom is selling AI by the token#

Token Factory is KT's bid to turn its billing operation into an AI business. The company already bills millions of phone and data customers; now it wants to bill AI tokens the same way, with corporate management functions due by December and an eventual expansion to individual developers. For Korean companies, the draw is practical: one contract, one won-denominated bill, and a router that steers routine work to cheaper models instead of defaulting to the most expensive one.

The open question is whether the 25% savings claim holds up in real workloads, and whether a telecom-run marketplace can keep pace with model releases that now land weekly. The 500-model target by year-end is the number to watch: if KT hits it, Token Factory becomes Korea's broadest single stop for model access; if the catalog stalls at 73, it is a useful procurement tool rather than a platform shift.

Sources: Aju Press; Seoul Economic Daily; MoneyToday; Superpower Daily.