LLM Token Counter

How to use this tool

Paste your prompt or document, choose Estimate tokens, then read the characters, words and estimated range. Copy the summary or clear the text.

Example

Input

Summarize the following release notes in three bullet points.

Output

Characters: 62
Words: 10
Estimated tokens: about 15 (range 12–19)

What does this tool do?

Language models read text as tokens, which are pieces of words, punctuation and whitespace chosen by a model-specific tokenizer. This counter reports characters, words and an estimated token count, plus a plausible range. The estimate uses a general rule of thumb rather than any vendor's tokenizer, so it is useful for sizing and budgeting but not for enforcing a hard limit.

Common mistakes and limitations

This is an approximation, not an exact token count. Real tokenizers differ by model and version, split code, numbers and non-English text differently, and add special tokens or message formatting that this page cannot see. Leave headroom below a context limit, and use your provider's own token-counting feature when an exact figure matters.

Tokens, context windows and input versus output

A token is a unit a model reads or writes: often a whole short word, part of a longer word, a number fragment or a punctuation mark. A context window is the total token budget for one request, and it normally has to hold your instructions, any supplied documents and the reply. Many APIs count input tokens and output tokens separately, so a long prompt and a long answer both consume budget and both can be billed.

Related: AI API Cost Calculator

Why two tokenizers give different counts

Each model family trains its own tokenizer, with its own vocabulary and splitting rules. The same sentence can therefore cost a different number of tokens in different models, and code, URLs, emoji and languages such as Japanese or Hindi often split into more tokens than plain English prose. That is why this page reports an estimate and a range. Use it to compare drafts and to leave headroom, then confirm with the provider's token counter before relying on a limit.

Same text, different tokenizers:
"Flutters.in tools" -> tokenizer A: 5 tokens
"Flutters.in tools" -> tokenizer B: 6 tokens

Planning a prompt that fits

Count the system prompt, the user input and any retrieved passages, then reserve room for the reply. If the total is too large, shorten instructions first, then split long source text into smaller passages.

Related: RAG Text Chunker, Prompt Template Builder, Word and Character Counter

Frequently asked questions