We use cookies for analytics and personalized ads. We use cookies for basic analytics and, where we show ads, to personalize them. Privacy PolicyRead our Privacy Policy.
Turn a token budget into a plain-language word count, or a word count into a rough token estimate — including 1 million tokens to words.
Quick answer: 1 token ≈ 0.75 words, so 1 million tokens ≈ 750,000 words (about 665,000-770,000 for English prose). 8,192 tokens ≈ 6,100 words, and 1 million characters ≈ 250,000 tokens.
Last updated October 1, 2026
Count Tokens LocallyScope: this gives a planning-grade range, not an exact count. For ordinary English prose, 1,000 tokens usually lands somewhere around 665-770 words — close enough to sketch a document's size, not close enough to bill against. Use the text estimator below for a closer figure on your own text, and the Token & Cost Calculator when you need an exact number.
Pick a direction, enter a number, and choose the text type that best matches what you're planning.
1,000,000 tokens using english prose (≈1.3-1.5 tokens per word)
At an average silent reading speed of ~225 words/minute, that's roughly 3,191 minutes of reading — handy for sizing a summary, brief, or report before you write it.
Nothing is uploaded. The text stays in your browser, and the estimate uses the text type selected in the converter above (english prose).
This is still an estimate. Real tokenizers differ by a few percent between providers.
For English text, one token is about three-quarters of a word, or about four characters. These four conversions cover almost every case.
The exact constants are 0.75 words per token, about 1.33 tokens per word, and about 4 characters per token. For code, JSON, or non-English text, multiply the token figure upward (see the text types above).
About 165,000 to 200,000 words, or about 250,000 tokens. An English word averages roughly 5-6 characters once you count the space after it.
| Characters | Estimated words | Estimated tokens |
|---|---|---|
| 1,000 | 167 – 200 | 250 |
| 10,000 | 1,667 – 2,000 | 2,500 |
| 100,000 | 16,667 – 20,000 | 25,000 |
| 1,000,000 | 166,667 – 200,000 | 250,000 |
Rough English-prose estimates for planning — the english prose setting above changes these ranges.
| Token budget | Estimated words |
|---|---|
| 1,000 tokens | 667 – 769 |
| 4,096 tokens | 2,731 – 3,151 |
| 8,192 tokens | 5,461 – 6,302 |
| 16,384 tokens | 10,923 – 12,603 |
| 32,768 tokens | 21,845 – 25,206 |
| 128,000 tokens | 85,333 – 98,462 |
| 200,000 tokens | 133,333 – 153,846 |
| 1,000,000 tokens | 666,667 – 769,231 |
The tool above gives you a range in one click. The harder part is knowing which direction to run it, which text type to pick, and when to stop trusting the estimate. Here's exactly that.
A token is the actual unit an AI model reads, generates, and gets billed in — not a word, not a character, but a sub-word chunk the model's tokenizer decides on. "Butterfly" might be one token or three depending on the tokenizer; a single emoji or an unusual Unicode character can quietly cost several tokens on its own. That's the entire reason this page deals in ranges: "words" is a convenience for humans, "tokens" is what the model and the invoice actually count.
Use the swap button to choose tokens → words (you have a token number and want words) or words → tokens (you have a draft or word count and want a token estimate).
Type your own figure, or tap one of the preset budgets (1K, 10K, 100K, 1M) to see a common reference point instantly.
Prose for ordinary writing, dense/multilingual for jargon-heavy or non-English text and text with a lot of non-ASCII characters or emoji, code/JSON for anything with braces, indentation, or repeated symbols — picking the wrong one is the single biggest source of a bad estimate.
The low-high spread is the honest answer. Use the copy button to paste the figure into a brief, or the embed snippet further down to drop the whole tool onto your own page.
This converter is deliberately approximate. For a real, billable token count on actual text, use the Token & Cost Calculator. If your question is specifically about how Claude counts and prices tokens — a common one, since Claude's tokenizer and context limits differ from OpenAI's — the Claude Models & Cost Guide covers that directly, including how to check how many tokens you have left on a Claude plan.
It's tempting to want one clean number — "a token is X words" — because it would make mental math easy. Real tokenizers don't work that way. They split text into sub-word pieces based on how frequently patterns appear in the data they were trained on, so the same idea written two different ways can tokenize into a noticeably different count even though the word count barely changes. This tool gives you a working range instead of a false-precision single number, because that range is the honest answer: close enough to plan with, wide enough to admit what it doesn't know.
Ordinary sentences with common words tokenize efficiently — most whole words map to one or two tokens, which is why the roughly 1.3-1.5 tokens-per-word range holds up well for articles, emails, and everyday writing.
Text that mixes languages, uses uncommon words, or is heavy with punctuation and formatting tends to split into more, smaller pieces — pushing the ratio up toward 1.5-2.0 tokens per word or higher.
Braces, brackets, indentation, and short repeated identifiers rarely match the whole-word patterns a tokenizer learned from natural language, which is why code commonly runs 2-3 tokens per space-separated 'word' or more.
For English writing, a useful planning range is about 1.3 to 1.5 tokens per word. It's close enough for early sizing — sketching how long a document needs to be, or roughly how much a token budget can hold — but it is not a substitute for actually counting tokens once you have real text to measure. Ratios also drift slightly across different providers' tokenizers, so two counters can legitimately disagree by a few percent on the exact same input.
Early content planning, briefing a writer on roughly how long a piece should be, or getting a gut-check on whether a document "feels" like it'll fit inside a rough context budget before you've written a single line.
Anything tied to an actual API call — checking a prompt fits inside a context window, estimating a bill, or setting an output cap. Providers charge and enforce limits in tokens, not words, so the final check always has to happen in tokens.
Say a team needs to summarize a batch of reports and has budgeted 500,000 tokens of input for the job across all documents. Using the English-prose range of 1.3-1.5 tokens per word, that budget can hold roughly 333,000 to 385,000 words of source material — somewhere around 650-770 pages at a typical 500 words per page. If the actual reports are dense, jargon-heavy technical documents rather than plain prose, the safer planning move is to use the "Dense or multilingual" setting instead, which brings the usable word count down to a more conservative 250,000-333,000 words for the same token budget — a meaningful difference to know before committing to a batch size.
Drop a lightweight version of this tool on your own site — self-contained, dark theme, links back to the live page.
<iframe src="https://apextechjournal.com/embed/tokens-to-words?ref=widget" width="100%" height="380" style="border:1px solid #262e39;border-radius:12px;" loading="lazy" title="Tokens to Words Converter by Apex Tech Journal" ></iframe>
Using the common rule of thumb (1 token ≈ 0.75 words), 1 million tokens is about 750,000 words. For plain English prose the realistic range is roughly 665,000 to 770,000 words, based on the 1.3-1.5 tokens-per-word range most tokenizers produce. Heavy code, JSON, or non-English text can push the actual figure well below that range.