Prompt Token Estimator
Estimate how many tokens your prompt uses for models like GPT and Claude, so you stay under context limits and control cost.
Estimate how many tokens a prompt will use before you send it to a large language model. Tokens — not words — are how models like GPT and Claude measure input and output, and they drive both context limits and cost. Paste your prompt to get a fast estimate along with word and character counts.
This gives a close approximation using standard heuristics; exact counts depend on each model's specific tokenizer, but the estimate is reliable for planning.
What a token is
A token is a chunk of text a model reads as a single unit — often a word, but frequently a piece of one. Common words are usually one token; longer or rarer words split into several. As a rough rule of thumb for English, one token is about four characters, or roughly ¾ of a word, so 1,000 tokens is around 750 words.
Why token counts matter
Two reasons: context limits and cost. Every model has a maximum number of tokens it can handle in one request; exceed it and your prompt is truncated or rejected. And API pricing is per token for both what you send and what you get back, so trimming a bloated prompt directly lowers your bill. Estimating before you send helps you fit the window and avoid surprises.
Trimming a prompt
To reduce tokens without losing meaning, cut filler and repetition, replace long examples with shorter ones, and remove restated instructions. Structured, concise prompts usually perform better and cost less. Watch the estimate update as you edit to find the leanest version that still works.
Budgeting a long prompt
When you're stuffing a document, chat history, and instructions into one request, tokens add up quickly and it's easy to overrun the context window. Estimating first tells you whether everything fits and roughly what the call will cost. If you're close to the limit, trim the least useful material — old conversation turns, verbose examples, restated instructions — and re-check. The context-usage readout here shows your prompt as a share of common window sizes so you can see at a glance how much headroom is left for the model's reply.
Why code and other languages cost more
The four-characters-per-token rule of thumb holds for ordinary English prose. It breaks down for code, which is full of punctuation and rare identifiers that split into many tokens, and for languages that don't use the Latin alphabet, where a single character can become several tokens. Treat this estimate as a reliable planning figure for English text and as a conservative lower bound for code and multilingual content.
Frequently asked questions
How accurate is the token estimate?
It uses standard heuristics (roughly four characters per token) and is close for typical English text. Exact counts vary by each model's tokenizer, especially for code, other languages, or unusual formatting.
How many words is 1,000 tokens?
Around 750 words of English on average, though it depends on the specific words. Code and non-English text often use more tokens per word.
Do output tokens count too?
Yes. Most APIs charge for both input and output tokens, so the model's reply adds to your total cost and must also fit within the context window.
Why not just count words?
Because models bill and budget by tokens, not words. A word can be one token or several, so word count alone can mislead you about cost and whether your prompt fits the context window.
Last updated: 2026-01-15