nowable.tech
// token economics

The token tax

Loading tokenizer…

Danish
English (same content)
Method: Tokenisation is real, computed in your browser with the GPT o200k tokenizer (the same family as GPT-4o and GPT-5). You can verify the numbers at tiktokenizer.vercel.app. GPT is among the most efficient tokenizers on European languages, so the figure here is a floor. Claude and DeepSeek split Danish even harder, up to 70 percent more tokens in nowable.tech's own benchmark, June 2026. The phenomenon is documented in Petrov et al. 2023, Ahia et al. 2023 and Ovcharov 2026. The token count is the same whether the text is sent to the model or written by it, so the tax lands on both ends of a conversation.

A nowable.tech tool.