0 XP

Tokenisation · Why words split

Frequency decides

Common words stay whole

A model reads text as tokens — chunks that are sometimes whole words and often fragments. The vocabulary was built by counting: text that appears constantly earned its own token, and anything rarer has to be assembled from smaller pieces. What matters is how often a word shows up in writing, not how long it is.

the3086cat9059sat10139
3 tokens: the / cat / sat

3 tokens · o200k_base

ant493idis129901est376ablishment160388arian21203ism2367
6 tokens: ant / idis / est / ablishment / arian / ism

6 tokens · o200k_base

Counts come from the real o200k_base vocabulary — the one GPT-4o reads with. Another model will split the same text differently. tiktoken