Token basics
How many tokens is 1,000 words?
A useful planning estimate is 1,000 English words ≈ 1,250–1,500 tokens. It is a range, not a conversion rule. The tokenizer reads pieces of text, so two documents with the same word count can consume different amounts of context.
A practical reference
For ordinary English prose, one token is often around three to four characters or roughly three-quarters of a word. That makes 750 words about 1,000 tokens and 1,000 words about 1,250–1,500 tokens.
Treat that estimate as planning shorthand. Count the actual material before an important API request or a long chat handoff.
- 500 English words: often 625–750 tokens
- 1,000 English words: often 1,250–1,500 tokens
- 10,000 English words: often 12,500–15,000 tokens
Why the ratio changes
Tokenizers split common text efficiently and unusual text less efficiently. Source code, identifiers, tables, repeated punctuation, emoji, and languages without spaces can produce a very different token-to-word ratio.
The surrounding chat matters too. Your prompt is only one part of the context alongside system instructions, previous turns, tool definitions, attachments, and the model's response.
When an estimate is enough
Use a word-based estimate for rough planning. Use a tokenizer when you are close to a limit, comparing request cost, splitting a document, or preparing a reproducible handoff.