There are plenty of token calculators online. All of them assume English. The same content in Korean spends more tokens, so a budget built on those numbers does not hold for a Korean-language service. We worked out how far off it lands, and how that differs by model, across the 194 models with published per-token pricing.
First: the multiplier is not something we measured
Counting exactly requires the model's tokenizer. OpenAI published o200k, but Anthropic and Google did not and will only count through their own APIs. So we approximate by character class: roughly four Latin characters per token, and about 0.67 tokens per Korean syllable. On that approximation Korean runs 2.7x for the same character count. Expect to be off by around 25%.
The multiplier also does not vary by model. Input and output inflate by the same factor, so all 194 land on 2.67x. Re-ranking on Korean changes nothing. None of this is news.
What differs is the amount
Say you send 10,000 Korean characters and get 2,000 back, a thousand times a month — summarising or editing a document, around thirty times a day. Against the same job in English, the extra you pay for Korean has a median of ₩4,109 a month. But the range runs from ₩148 to ₩1,540,831, a spread of more than ten thousand times.
| Model | Rate (in/out) | Korean | English | Extra |
|---|---|---|---|---|
| o1-pro | $150 / $600 | ₩2,465,453 | ₩924,622 | ₩1,540,831 |
| gpt-4 | $30 / $60 | ₩383,533 | ₩143,830 | ₩239,703 |
| qwen3.6-plus | $0.33 / $1.95 | ₩6,574 | ₩2,466 | ₩4,109 |
| mistral-nemo | $0.02 / $0.03 | ₩237 | ₩89 | ₩148 |
o1-pro alone eats an extra ₩1.54M a month for being Korean. mistral-nemo pays ₩148 more for the same reason. Both sit at the same 2.67x.
So the choice of model weighs more
This is what actually changes when you build in Korean. Not the multiplier, the amplification. A choice between two models that looks like a ₩30,000 monthly difference in English becomes ₩80,000 in Korean. The cost of choosing wrong grows by 2.7x.
Output especially. Korean inflates the reply by the same factor, and output rates run 4.1x input at the median. In the scenario above, 45% of the extra comes from output, rising to 71% on output-heavy models. The habit of picking by input rate alone costs more in Korean.
How to check
The numbers above are one scenario. Different volumes give different conclusions. Paste the prompt you actually send into the calculator and it will line the models up at your own volume. Pasted text never goes into the address, so the result link is safe to share.
Each model page also carries the same scenario, Korean and English side by side.