Token Counter: Words, Characters & Tokens
Count tokens, words and characters for any text, and convert between them.
With the default inputs, ~1,333 tokens — 1,000 words · 5,332 characters. Enter your own numbers below to recompute instantly; the full step-by-step math is shown under the worked example.
Results update automatically as you type.
- Tokens
- 1,333
- Words
- 1,000
- Characters
- 5,332
- Input cost at this model(GPT-4o)
- $0.00333
| Prose / English | 4 | 1,333 |
| JSON | 3.5 | 1,523 |
| Code | 3.3 | 1,616 |
| Non-Latin / other | 2 | 2,666 |
Paste text to see exactly how many tokens, words, characters and sentences it contains, and what it costs as input at any model's live price. Leave the box empty and enter a number instead to convert straight between words, characters and tokens — useful when you're sizing a document you haven't written yet. Either way you also get the same length priced across every content type, because prose, JSON, code and non-Latin scripts turn characters into tokens at very different rates.
How this is calculated
Paste text and we count it directly: characters, characters without spaces, words and sentences are exact, and the token count comes from the real o200k_base tokenizer running in your browser — no upload, nothing leaves the page. With the text box empty we switch to conversion mode and derive words, characters and tokens from whichever single number you enter, using content-type ratios (English prose ≈ 0.75 words and ≈4 characters per token). Either way the token count is priced as input at your chosen model's live rate, and the table re-runs the same length at every content-type ratio so you can see what format you're paying for.
Is this a good result? What to do next
A pasted count is exact for OpenAI-family models and within a few percent for others; converted estimates are planning figures, not billing guarantees. If a number looks surprisingly high, check the content type first — code, JSON and non-Latin scripts pack far more tokens into the same characters, and that single setting moves the estimate more than anything else on the page.
Typical planning ranges
- English prose
- ~4 characters per token
- Code
- ~3.3 characters per token
- JSON
- ~3.5 characters per token
- 1,000 words of prose
- ~1,333 tokens
- Tweet length (280 chars)
- ~70 tokens
Ranges are typical planning figures to sanity-check your result, not authoritative benchmarks. Your numbers will vary with use case, volume, and vendor.
How to improve this number
- Trim boilerplate and repeated instructions — system prompts are resent on every call.
- Prefer prose or a compact format over verbose JSON when structure isn't required.
- Strip whitespace and repeated keys from structured payloads before sending them.
- Pick the content type that matches your text for a closer estimate.
Common mistakes
- Assuming one word equals one token; English averages ~0.75 words per token.
- Applying prose ratios to code or JSON, which are denser.
- Counting the prompt once when a chat resends the whole history on every turn.
- Treating an estimate as exact for billing rather than planning.
When to use a different approach
To check whether a document fits a model in one call, use the will-it-fit context checker. To price full traffic rather than a single prompt, use the LLM API cost calculator. To see how many tokens a dollar buys across models, use the tokens-per-dollar calculator.
Worked example (defaults)
With the default inputs above, here is the result:
- Tokens
- 1,333
- Words
- 1,000
- Characters
- 5,332
- Input cost at this model(GPT-4o)
- $0.00333
| Prose / English | 4 | 1,333 |
| JSON | 3.5 | 1,523 |
| Code | 3.3 | 1,616 |
| Non-Latin / other | 2 | 2,666 |
- Ratios: 1 token ~= 0.75 words ~= ~4 characters (prose).
- 1,000 words -> ~1,333 tokens
Sources & references
Frequently asked questions
How accurate is this token counter?+
When you paste text, the count is exact — it runs the real o200k_base tokenizer in your browser, the same family OpenAI's current models use. The converted estimates (and counts for other model families) use well-established character-to-token ratios and land within a few percent for typical text.
How many tokens is a word?+
For English prose, one token is roughly 0.75 words, so 1,000 words is about 1,333 tokens and a typical 500-word page is around 650–700 tokens. Code, JSON and non-Latin text are denser, so the same word count costs more tokens.
How many characters are in a token?+
About 4 characters per token for English prose, ~3.5 for JSON, ~3.3 for code, and as few as ~2 for non-Latin scripts. A 280-character tweet is roughly 70 tokens.
Why do token counts differ between models?+
Each model family ships its own tokenizer, so the same text maps to slightly different counts across OpenAI, Claude and Gemini — usually within a few percent. Switch the model above to price your text at whichever one you're actually calling.
Why does code or JSON use more tokens than prose?+
Braces, quotes, commas, indentation and repeated keys all become tokens of their own, so structured text packs fewer characters into each token. The comparison table shows the same content priced at every ratio, which is the quickest way to see what reformatting a payload would save.
Does this count my prompt or the model's reply?+
It counts whatever you paste, and prices it as input. Output tokens are billed at a separate, usually higher rate — use the output token cost calculator for the response side.
Related calculators
Answering a real question?
This calculator powers these problem-solving guides: