Tokens in French

What the numbers say

The same website copy is 200 tokens in English and 244 tokens in French on the newer o200k_base encoding, a ratio of 1.22×. On the older cl100k_base encoding the French copy needs 278 tokens, which is 1.39× the English count of 200. So on the newer encoding French costs only a modest amount more than English. Among the 29 non-English languages we measured, French ranks number 10 from the cheapest.

The newer encoding made a real difference: o200k_base used 12% fewer tokens than cl100k_base for French, compared with 34% across all the non-English languages in the data. That is less than average, so French gained less from the newer vocabulary than most.

Why French? French uses the Latin alphabet with accents (é, è, ê, à, ç) and apostrophes joining short words (l’, d’, qu’). It is one of the best-represented languages in training text, so common words and accented endings are often single tokens.

Languages written in the Latin alphabet share the letters English text uses, which are the best-represented in training data, so words are usually covered by a few tokens. Accented letters add a little, and long compound or suffixed words add more, but the cost stays within a modest multiple of English.

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 French prompt costs about $1.22, and about $1.39 on cl100k_base. A document of 1,000 French characters is roughly 226 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 French text

1. Use a model with the newer tokenizer. For French, o200k_base used 12% 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 French; 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.