🔤 AI & text

Token to Word Converter

Estimate how many words, characters, sentences or pages a token budget represents. Enter a value, choose the units, then press Convert.

Written and reviewed by SM Arijul Islam · Last reviewed 29 October 2026 · Editorial standards · Report an error

About the Token to Word Converter

Language models do not read words; they read tokens. For ordinary English prose, one token is roughly four characters or three-quarters of a word, so 1,000 tokens is about 750 words. That rule is a planning aid, not a guarantee: code, URLs, numbers, emoji and non-Latin scripts tokenise very differently.

The converter also estimates pages at 500 words per page and sentences at 20 tokens each. Use it to sanity-check a prompt or a context-window limit, then measure with the tokenizer of the model you actually use if the exact count matters.

How the calculation works

1 token ≈ 4 characters of English text.
1 token ≈ 0.75 words; 1 word ≈ 1.333 tokens.
1 page ≈ 500 words ≈ 667 tokens.
1 sentence ≈ 20 tokens (very rough).

Reference values

Typical English prose equivalents
Tokens	Words (approx.)	Characters	Pages
100	75	400	0.15
500	375	2,000	0.75
1,000	750	4,000	1.5
4,000	3,000	16,000	6
8,000	6,000	32,000	12

Worked example: fitting an article into a context window

A 2,000-word article is about 2,667 tokens under the four-characters-per-token rule, or roughly 10,700 characters. A 128,000-token context window therefore holds about 96,000 words of plain English, leaving room for the prompt and the response. Code and URLs tokenise less efficiently: a dense JSON file can use twice as many tokens as its word count suggests.

Treat the estimate as a planning figure. If you are close to a hard limit, measure with the tokeniser published by the model provider. The same paragraph can differ by 10–20% between tokenisers, and non-Latin scripts or emoji can differ by much more.

Turning tokens into cost

Application programming interfaces price input and output tokens separately, and output is usually several times more expensive. If a model charges $3 per million input tokens and $15 per million output, a 2,000-token prompt producing a 500-token answer costs 2,000 × 3 ÷ 1,000,000 = $0.006 plus 500 × 15 ÷ 1,000,000 = $0.0075, or $0.0135 in total. A million requests at that size would cost about $13,500, which is why prompt trimming matters at scale.

Long conversations also resend history with every turn. A 10-turn chat does not cost ten times the first message; it costs the sum of every growing context, so the total grows roughly quadratically. Trimming, summarising or caching the stable prefix are the usual ways to keep that in check.

Frequently asked questions

Is 1 token always 4 characters?

For English prose, roughly. JSON, code and punctuation-heavy text can use two or three characters per token; some non-Latin scripts use more tokens per character.

Can I use this for GPT context limits?

As a rough budget, yes. Leave headroom: a 4,000-token estimate can become 5,000 once you add formatting, code blocks or a system prompt.

How many pages is 10,000 tokens?

About 15 pages at 500 words per page, or roughly 7,500 words of ordinary prose.

Do all language models tokenise text the same way?

No. GPT, Claude, Gemini and open-weight models use different tokenisers, so the same text can produce different token counts. Use the provider's tokeniser when the exact count matters.

How do I estimate API cost from tokens?

Multiply input tokens by the input price per million and output tokens by the output price per million, then add them. Remember that chat history is resent each turn, so long conversations cost more than the sum of their individual messages.

Sources and standards

Token counts depend on the tokeniser a model uses; there is no universal words-to-tokens factor. The counts here follow OpenAI's tiktoken encoding, the reference implementation for its models, and the page states which encoding is assumed rather than presenting the ratio as a fixed rule.