How to use the token counter
- Paste your text — a prompt, document or conversation.
- Pick the model family you're using; the estimate adjusts to how that family tends to tokenize.
- Read the estimated tokens to check you're within the model's context window and to gauge API cost.
What is a token?
Language models don't read words — they read tokens, which are chunks of text roughly ¾ of a word on average in English. "Cat" is one token; "unbelievable" might be three. Models charge by tokens and have a maximum context window measured in tokens, so knowing your token count tells you both whether your text will fit and roughly what an API call will cost.
How accurate is this estimate?
This tool gives a fast, close estimate using the well-known rule that English averages about 4 characters per token, adjusted per model family. It is not the exact byte-pair tokenizer each provider uses, so treat it as a reliable ballpark (usually within a few percent for normal prose) rather than an exact billing figure. Code, other languages, and unusual formatting tokenize differently and may vary more.
Benefits
- Estimates prompt tokens for GPT, Claude and Gemini so you can stay inside context limits.
- Shows tokens alongside word and character counts for a full picture.
- Helps you gauge API cost before you send a prompt.
- Runs privately in your browser — prompts are never uploaded.
Limitations to know
- It is an estimate, not the exact byte-pair tokenizer each provider uses.
- Code, non-English text and unusual formatting tokenize differently and may vary more.
- It measures the text you paste, not a model’s response tokens.
Common mistakes to avoid
- Treating the estimate as an exact billing figure — it is a close ballpark.
- Forgetting that API cost includes the model’s output tokens too, not just your input.
- Assuming other languages or code tokenize at the same rate as English prose.
Alternatives
For plain word and character counts use the Word Counter; to shorten a prompt that is too long, try the AI Text Summarizer.
Last updated: August 2026 · Reviewed by the AI Toolbox editorial team.
Frequently asked questions
Is my text sent anywhere?
No — the estimate is calculated entirely in your browser. Nothing is uploaded.
Why does the count differ from OpenAI's?
Each provider uses its own byte-pair tokenizer. This tool estimates rather than replicating those exactly, so expect a small difference — close enough for planning prompts and budgets.
How many tokens is a page of text?
A typical single-spaced page (~500 words) is roughly 650–750 tokens.
Do input and output both count?
Yes — most APIs bill for prompt (input) tokens plus the model's response (output) tokens. This tool measures the text you paste.