Words to Tokens: Estimate Prompt Size Before You Paste
How to convert a word count into a token budget — the 1.3 rule, per-content corrections, and a converter that does it for you.
Updated Aug 8, 2026 · Reviewed by PromptCostLab
Token and pricing information changes frequently. Use this guide for planning and verify model-specific limits and prices with the provider before production billing.
The 1.3 rule
Since one token is about 0.75 English words, one word is about 1.3 tokens. That single multiplier converts any word count into a usable token estimate: a 500-word email is ~650 tokens; a 2,000-word article is ~2,600; a 60,000-word manuscript is ~78,000. Keep a 15–20% band around the result for style effects.
Why more tokens than words? Frequent short words compress to one token, but the average English word is long enough — and rare enough — that most split into pieces. Plurals, suffixes, and punctuation all add fragments.
Words to tokens table
Common document sizes converted:
- 100 words ≈ 130 tokens (120–155)
- 250 words ≈ 325 tokens — a concise email
- 500 words ≈ 650 tokens — a typical blog section
- 750 words ≈ 975 tokens — an essay page
- 1,000 words ≈ 1,300 tokens
- 2,000 words ≈ 2,600 tokens — a long article
- 5,000 words ≈ 6,500 tokens — a whitepaper
- 10,000 words ≈ 13,000 tokens — a short ebook
- 50,000 words ≈ 65,000 tokens — a novel
- 150,000 words ≈ 195,000 tokens — nearly fills a 200K window
Adjust for content type
The 1.3 multiplier assumes prose. Dense professional writing with jargon, citations, and numbers runs closer to 1.5 tokens per word. Source code — with its operators, indentation, and identifiers — runs 1.5–2. Non-Latin scripts can multiply several-fold: CJK text costs roughly one token per character, so a 1,000-'word' Japanese document is far more expensive than its English summary.
Dialogue-heavy chat transcripts sit nearer 1.4 because every speaker turn adds structural tokens. Markdown and JSON add overhead per symbol too.
From word count to dollar cost
The full chain: words × 1.3 gives tokens; tokens ÷ 1,000,000 × price gives cost. A 1,000-word prompt on Claude Sonnet 5 is ~1,300 tokens ≈ $0.0026 of input — tiny per call, but 3,000 calls a day turns it into ~$8 a month of input before output. At scale the multiplier's error band matters: 20% estimate error on a five-figure monthly bill is real money.
Paste the actual text into the PromptCostLab prompt calculator for a tokenizer-backed count and per-model cost comparison — it replaces the 1.3 guess with a real measurement while keeping your text local.
Estimating before the text exists
Words-to-tokens math earns its keep at planning time: sizing a knowledge base before indexing ('our 40M-word wiki is ~52M tokens — retrieval, not whole-corpus context'), scoping output budgets ('a 300-word answer is ~390 output tokens at $10–50 per million'), and choosing plans. When the text exists, measure it; when it doesn't, estimate with the multiplier and a safety band.
How many tokens is one word?
An English word costs about 1.3 tokens on average — the flip side of the 1 token ≈ 0.75 words rule. Common words are a single token; longer or rarer words split into multiple pieces, so dense technical prose runs higher, around 1.5 tokens per word.
How many tokens is a 1,000-word article?
Roughly 1,300 tokens for clean English (range 1,200–1,550). Code and non-English text run higher — assume 1.5–2 tokens per word for source code.
Is there an exact words-to-tokens formula?
No — it depends on the model family's tokenizer, the vocabulary, and your text. Multipliers like 1.3 are planning estimates. For a real number, tokenize the actual text; PromptCostLab's converter does it in your browser.
Prices, token behavior, and model limits change. These primary provider docs are the right place to verify the details before shipping.