Skip to content

AI GPT Token Counter

Text ToolsAI Writing Assistants
Estimate from

The context window or output cap you are working against.

Estimated tokens —
Words
—
Characters
—
Tokens by characters
—
Tokens by words
—
Share of limit
—
Tokens left
—

The counter estimates GPT tokens with OpenAI's rule of thumb: one token is roughly four characters of English, or three quarters of a word. A 75-word paragraph comes out near 100 tokens. Set your context window or output cap and the counter shows the share used and the tokens left.

About this tool

Prompt engineers, developers sizing API requests and writers pasting long documents into a chat model all hit the same question: will this fit, and roughly what will it cost in tokens? This counter gives a fast estimate from two published rules, characters divided by four and words multiplied by four thirds, and lets you pick either one or their average. It also compares the result with a token limit you set, such as a 128,000-token context window, and warns when the text runs over. The limit to keep in mind is that this is an estimate, not the model's real tokenizer: code, numbers, unusual words and non-Latin scripts usually split into more tokens than the rules predict, so leave headroom of 10 to 20 percent.

How to use it

  1. Add your text

    Paste the prompt or document into the box, or open a .txt file from your device.

  2. Pick a rule

    Choose characters ÷ 4, words × 4/3, or the average of the two.

  3. Set the limit

    Enter the context window or output cap you are working against.

  4. Read the estimate

    See estimated tokens, both rule results, the share of the limit used and tokens left.

Examples

75 words by the word rule

Prompt or text
one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three one two three
Estimate from
Words × 4/3
Token limit
128000

Result Estimated tokens: 100
Words: 75
Characters: 349
Tokens by characters: 88
Tokens by words: 100
Share of limit: 0.1%
Tokens left: 127,900

  1. Characters: 349 ÷ 4 = 88 tokens (rounded up).
  2. Words: 75 × 4 ÷ 3 = 100 tokens (rounded up).
  3. Using the word rule: 100 tokens.

OpenAI's guide puts 75 English words at about 100 tokens, which is what the word rule returns here.

A one-line pangram

Prompt or text
The quick brown fox jumps over the lazy dog.
Estimate from
Average of both
Token limit
128000

Result Estimated tokens: 12
Words: 9
Characters: 44
Tokens by characters: 11
Tokens by words: 12
Share of limit: 0%
Tokens left: 127,988

  1. Characters: 44 ÷ 4 = 11 tokens (rounded up).
  2. Words: 9 × 4 ÷ 3 = 12 tokens (rounded up).
  3. Average of both: (11 + 12) ÷ 2 = 12 tokens.

44 characters give 11 tokens and 9 words give 12; the average rounds up to 12.

Checking a tight cap

Prompt or text
The quick brown fox jumps over the lazy dog.
Estimate from
Characters ÷ 4
Token limit
10

Result Estimated tokens: 11
Words: 9
Characters: 44
Tokens by characters: 11
Tokens by words: 12
Share of limit: 110%
Tokens left: 0

  1. Characters: 44 ÷ 4 = 11 tokens (rounded up).
  2. Words: 9 × 4 ÷ 3 = 12 tokens (rounded up).
  3. Using the character rule: 11 tokens.

With a 10-token cap, the 11-token estimate is flagged as one token over.

How it is calculated

tokens ≈ characters ÷ 4, or words × 4 ÷ 3; average = (both) ÷ 2, rounded up

characters
visible characters including spaces and punctuation
words
runs of Unicode letters or digits
4
OpenAI's typical characters per token for English text

GPT models split text into byte-pair tokens: common words are a single token, rarer words break into pieces. OpenAI's guidance summarises this for English as about four characters or three quarters of a word per token. The two rules disagree on short words and long technical terms, so the average is the default. Every result is rounded up, since an underestimate is the costlier error when filling a context window.

When not to use it

  • When billing must match to the token, run the model's own tokenizer instead.
  • For source code, JSON or long numbers, which tokenise far less evenly than prose.
  • For Chinese, Japanese, Arabic or other non-Latin text, where tokens per character run much higher.

Common mistakes

  • Counting only the prompt and forgetting that the model's reply also uses the context window.
  • Leaving no headroom, so a prompt estimated just under the limit is rejected.
  • Ignoring system messages and chat formatting, which add tokens to every request.

Frequently asked questions

How many words is 1,000 tokens?

About 750 English words by OpenAI's rule of thumb. Plain conversational prose often comes a little higher, while technical writing with long terms comes lower.

Why is this an estimate rather than an exact count?

Exact counts need the model's byte-pair vocabulary, a list of roughly 100,000 to 200,000 pieces that differs between model families. The character and word rules give a close figure for English prose without loading that vocabulary.

Which rule should I pick?

The average suits most prose. Characters ÷ 4 tracks better for text with many long words or URLs; words × 4/3 suits short, plain sentences. If the two differ widely, the text is unusual and deserves extra headroom.

Do spaces and punctuation count as tokens?

Usually a space joins the word that follows it into one token, while punctuation marks are often tokens of their own. The character rule includes both, which is why it counts every space and full stop.

Does other-language text use more tokens?

Generally yes. Scripts outside the Latin alphabet often take one or more tokens per character, so the counter warns when most of the text is non-Latin and the estimate is likely low.

Is my text sent to an AI service?

No. The estimate is worked out in your browser from character and word counts, and the text is not passed to any model or server.

Cite this tool

Using this tool in an assignment, article or report? Copy a citation, or a link for your website.

APA

Tools Directory Editorial Team. (2026, October 25). AI GPT Token Counter. Tools Directory. Retrieved October 26, 2026, from https://tools.milunir.com/text-tools/ai-writing-assistants/ai-gpt-token-counter/

MLA

Tools Directory Editorial Team. “AI GPT Token Counter.” Tools Directory, 25 Oct. 2026, https://tools.milunir.com/text-tools/ai-writing-assistants/ai-gpt-token-counter/. Accessed 26 Oct. 2026.

Chicago

Tools Directory Editorial Team. “AI GPT Token Counter.” Tools Directory. Last modified October 25, 2026. Accessed October 26, 2026. https://tools.milunir.com/text-tools/ai-writing-assistants/ai-gpt-token-counter/.

Link<a href="https://tools.milunir.com/text-tools/ai-writing-assistants/ai-gpt-token-counter/">AI GPT Token Counter – Tools Directory</a>