Tokens in Greek

What the numbers say

The same website copy is 200 tokens in English and 386 tokens in Greek on the newer o200k_base encoding, a ratio of 1.93×. On the older cl100k_base encoding the Greek copy needs 924 tokens, which is 4.62× the English count of 200. At the same price per token, the same message costs about 1.93× as much in Greek on the newer encoding. Among the 29 non-English languages we measured, Greek ranks number 29 from the cheapest.

The newer encoding made a real difference: o200k_base used 58% fewer tokens than cl100k_base for Greek, compared with 34% across all the non-English languages in the data. That is more than average, so Greek gained more from the newer vocabulary than most.

Why Greek? Greek has its own alphabet, with accent marks (tonos) on stressed vowels and a rich set of word endings. Each Greek letter is two bytes in UTF-8, and Greek is a small share of the text many tokenizers were trained on, so even common words are often cut into several pieces.

The Greek alphabet is used by one main language, so it has a small share of the text a tokenizer learns from. Letters are two bytes each, and with accents and long inflected endings, a word needs more pieces than the same word in a Latin-script language. Newer vocabularies include more Greek, which is why the improvement from the older to the newer encoding is large.

What it means for your bill

API prices are per token, so the ratio is also a price ratio. If a prompt costs $1.00 in English on o200k_base, the equivalent Greek prompt costs about $1.93, and about $4.62 on cl100k_base. A document of 1,000 Greek characters is roughly 354 tokens on o200k_base. These are estimates from one sample: translated text is never exactly the same length as the original, and your text may be formal, technical or full of names, which changes the ratio.

To turn tokens into money, enter your provider's input and output prices per million tokens in the token calculator. For how pricing works, see how LLM API pricing works.

How to use fewer tokens with Greek text

1. Use a model with the newer tokenizer. For Greek, o200k_base used 58% fewer tokens than cl100k_base in this sample, so a GPT-4o-class model needs fewer tokens than an older one for the same text.
2. Count before you send. Paste a typical message into the token calculator to see the real count, then multiply by the number of requests.
3. Keep fixed instructions short. A system prompt that is sent with every request is paid for every time. Writing the standing instructions in English is often cheaper per token than in Greek; test that answers stay as good as you need.
4. Cut what is not needed. Remove greetings, repeated context and long examples. The guide to reducing token usage has more.
5. Be careful with emoji, rare characters and unusual formatting. They are often split into several byte-level tokens each.