The unit models actually process.
WHAT A TOKEN IS
A fragment of text. Roughly a word, sometimes part of one, sometimes punctuation.
Common words are single tokens. Unusual words split into several.
WHY IT MATTERS
Context limits are measured in tokens Pricing for API access is per token Response length limits are in tokens
A ROUGH GUIDE
For English, one token is approximately three-quarters of a word.
A thousand words is roughly 1,300 tokens.
WHY THE RATIO VARIES
Languages other than English frequently use more tokens per word Technical terms and unusual names split into more Code tokenises differently from prose
WHAT THIS MEANS FOR COST
A long prompt costs more than a short one, every time it is sent.
In a system sending the same lengthy instructions repeatedly, that adds up.
PRACTICAL IMPLICATIONS
Concise prompts cost less Long documents may need splitting Output limits are why responses sometimes stop mid-sentence
WHEN YOU NEED TO THINK ABOUT IT
Building anything with an API. For ordinary chat use, rarely.
