KT Token Factory Is Live: 73 AI Models on One Platform, Won Billing, and a 30% Discount With a Catch — Explained Honestly
Nothing here is sponsored — no affiliate links on this page. Every fact below comes from Korean reporting of the 30 October 2026 launch (Aju Press, NoCut News, Financial News) and KT's 27 October press briefing as reported by Asia Business Daily. We have not tested Token Factory ourselves, and where KT's numbers are claims rather than verified results, we label them as claims.
South Korean telecom giant KT has officially launched Token Factory, its AI transformation (AX) platform, today — 30 October 2026. The pitch is simple and, for companies drowning in AI subscriptions, genuinely appealing: pick from dozens of AI models on one platform, watch all your usage in one place, and pay one bill in Korean won. At launch that means 73 models, including OpenAI's GPT series, Anthropic's Claude, Mistral, DeepSeek, xAI and Alibaba Cloud — plus a 30% discount on seven models served from KT's own domestic infrastructure. But read the fine print before getting excited: this is a corporate-only launch, the discount has a verification gate and a 30 November expiry, and the headline "500+ models" number is not what you get today.
🔍 How we put this together
We used Aju Press, NoCut News and Financial News reporting of the 30 October 2026 launch, and Asia Business Daily's report of KT's 27 October press briefing in Seoul for the router, OpenRouter partnership and roadmap details. Where a number is KT's claim (the 25% token saving) or a future target (500+ models in November), we say so — we do not present either as a tested fact.What actually launched
One platform, 73 models. Token Factory lets companies select and use a range of AI models from a single platform, managing everything from model selection and usage monitoring to billing in one place — no separate contracts for each model provider. KT plans 73 domestic and international models at launch: 7 served directly on KT's domestic infrastructure and GPUs, plus 66 global models.
The global catalogue covers the names you know. The 66 global models include OpenAI's GPT series, Anthropic's Claude, Mistral, DeepSeek, xAI and Alibaba Cloud. That breadth is the point: different teams inside one company can use different models without each team running its own subscription and expense trail.
Billing in Korean won. Usage fees are billed in won, using the billing capability KT built up through its telecom business. For Korean companies, that removes the currency-conversion friction of paying multiple overseas providers in dollars — and, KT says, customers pay only for the tokens they actually use.
KT says it will not store customer data or use it for training. That commitment matters for companies worried about data sovereignty — the fear that relying on overseas AI services means losing control of their data. KT's answer is domestic inference: the seven directly-served models run from KT data centres in Korea. Note this is KT's stated policy; it is not something we have independently audited.
The 30% discount — and its catch
To mark the launch, KT is discounting usage fees by 30% for the seven models it provides directly through domestic infrastructure, until 30 November 2026. The seven are:
- Mi:dm K 3.0 Pro
- Solar-Pro 4
- Motif 3
- NVIDIA Nemotron 3 Ultra 550B A55B
- GLM 5.3 Flash
- Kimi K3
- Gemma 4 31B
The catch is the gate: the discount applies to corporate customers who complete corporate verification and sign a product contract. It is a one-month launch promotion, not a permanent price cut — and it covers only those seven domestic models, not the 66 global ones. If you were hoping for 30% off GPT or Claude, that is not what this is.
Auto Model Router: the feature to watch
KT's proprietary Auto Model Router analyses the intent and complexity of a query in real time, sending tasks that need advanced reasoning to premium models and simple jobs like summarisation and translation to cheaper models. Combined with prompt-compression technology that strips unnecessary repetition from inputs, KT says the system can reduce token usage for operating AI services by up to 25%.
That "up to 25%" deserves a clear label: it is KT's claim, not an independently verified test. The idea is sound — routing easy work to cheap models is one of the oldest cost-saving tricks in production AI — but no outside benchmark in the reporting confirms 25% on real workloads. Treat it as a target to test in your own usage, not a guaranteed saving.
The OpenRouter partnership — and the 500+ number
KT's partnership with OpenRouter, the global AI model aggregation platform, is a first of its kind in South Korea. It is how the catalogue grows: KT plans to expand from 73 models to at least 500 models in November by linking up with OpenRouter, and also plans to distribute domestic models overseas through OpenRouter in the other direction.
Be precise about the timeline: 500+ is a November expansion target, not day-one availability. Headlines that say Token Factory "offers 500+ models" today are conflating the launch catalogue with the November plan. What you can actually use on launch day is 73.
Who this is for — and who should wait
- Korean companies juggling multiple AI subscriptions: you are the launch audience. One platform, one bill in won, usage visibility across teams, and a month of 30% off the domestic models if you complete verification and contract.
- Teams worried about data sovereignty: domestic inference for the seven models, plus KT's no-storage, no-training commitment, is a real argument — subject to the usual "verify the contract wording" advice.
- Individual developers and non-Korean users: wait. Token Factory is B2B-only at launch, with a corporate verification gate and won billing. KT's stated roadmap is corporate management functions by December, then individual/developer access with subscription pricing by March next year.
- Anyone chasing the 500-model catalogue: wait until November, then check what actually arrived — the OpenRouter expansion is a plan, and plans slip.
What we cannot verify — stated plainly
- The 25% token saving: KT's claim via its Auto Model Router and prompt compression. No independent test in the reporting confirms it.
- The 500+ November catalogue: a target tied to the OpenRouter link-up, not a launched fact.
- Real-world pricing beyond the discount: no per-model price list is published in the launch reporting, so we state no prices here. The only price fact is the 30% discount on the seven domestic models until 30 November.
- Individual/developer access and subscription pricing: planned by March next year, not available now, with no prices announced.
The honest summary is this: Token Factory is a sensible, genuinely useful idea for Korean companies — one platform, one won bill, 73 models on day one — with a launch discount that is real but narrow, and two headline numbers (25% savings, 500+ models) that are a claim and a target respectively, not facts you can bank on today. If the November expansion lands and independent users confirm the routing savings, this becomes a much bigger story. Until then, it is a strong corporate launch with honest limits — and that is worth saying plainly.
Sources: Aju Press, "KT Launches 'Token Factory' with 30% Discount on AI Model Usage Fees", 30 October 2026, https://www.ajupress.com/view/20261030090870685 ; NoCut News, "KT Opens Token Factory, Discounts Fees for 7 AI Models by 30%", 30 October 2026, https://en.nocutnews.co.kr/news/6599801 ; Financial News, "KT to cut fees for seven AI models by 30% on its Token Factory platform", 30 October 2026, https://en.fnnews.com/news/202610300914106377 ; Asia Business Daily, "KT's Token Factory Offers 500+ AI Models, Cuts Token Usage by 25%", 27 October 2026 (KT press briefing, Seoul), https://www.asiae.co.kr/en/article/IT/2026102709300050978 .