Token estimation guide
Tokens to Words: Conversion Guide
For ordinary English prose, 100 tokens is roughly 75 words, and 1,000 words is roughly 1,333 tokens. Treat that as a planning estimate: the exact result changes with the model, language, punctuation, code, and formatting.
Quick token-to-word conversion
Use these formulas for a first-pass estimate of English prose:
- Words ≈ tokens × 0.75
- Tokens ≈ words ÷ 0.75
Common estimates
100 tokens ≈ 75 words
500 tokens ≈ 375 words
1,000 tokens ≈ 750 words
2,000 tokens ≈ 1,500 words
10,000 tokens ≈ 7,500 words
Words-to-tokens examples
A 500-word article is roughly 667 tokens. A 1,500-word report is roughly 2,000 tokens. A 10,000-word document is roughly 13,333 tokens. These figures estimate the document text only; a real API request may also include system instructions, chat history, retrieved context, and tool results.
Why exact token counts vary
Models do not read text one word at a time. Their tokenizers split text into reusable pieces that may be a whole word, part of a word, punctuation, or whitespace. A common word may fit in one token, while an unfamiliar name, long number, URL, or symbol-heavy string may take several.
Code and structured data often use more tokens per visible word because punctuation, indentation, property names, and repeated syntax all count. Languages that do not separate words with spaces can also differ substantially from the English rule of thumb. For Korean, Japanese, mixed-language text, or production billing, test a representative sample with the provider's tokenizer or token-counting API.
Input and output both count
Converting a document to tokens estimates only part of a workflow. API billing typically separates input tokens and output tokens. Add the answer length, repeated workflow rounds, retries, and daily usage before calculating a monthly budget.
Use an estimate or an exact count?
Use the conversion formula for early planning, content limits, and rough budget ranges. Use an exact provider count before enforcing a context limit or approving a production budget. OpenAI offers an interactive tokenizer, and Gemini documents a token-counting method because tokenization depends on the actual request.
Official references: OpenAI Tokenizer and Gemini token counting guide.