Token
Convert.
Count, inspect and compare AI tokens, then estimate what your prompt could cost across OpenAI models. Fast, private, and entirely browser-based.
Choose the tokenizer encoding used to split and count your text. Model selection for API pricing is separate below.
Large input - tokenization may take a moment.
Each colored block is one token. Hover a token for its ID, or click it to inspect the piece and position. Spaces, tabs, and line breaks inside tokens are shown as visible marks.
Your text stays in your browser - it is never uploaded, logged, or sent to analytics.
Estimated API Cost
Pricing is indicative and may change. Verify with the provider before making financial decisions.
Compare Tokenizers
How Token Counting Works
Token counting works by running your text through the same subword tokenizer a language model uses internally. Instead of splitting text into whole words, the tokenizer breaks it into smaller, reusable pieces - sometimes a full word, sometimes just part of one - based on patterns learned from large amounts of text.
TokenConvert runs OpenAI-compatible tokenizers locally in your browser. When the selected encoding matches the model's tokenizer, the resulting token count reflects that encoding accurately. Different model families may tokenize the same text differently.
What Is a Token?
A token is the basic unit a language model reads and generates text in - not quite a word, not quite a character. Common words are often a single token, while rarer words, numbers, and punctuation are frequently split into two or more tokens.
Why Token Count Matters
Every model has a maximum context window measured in tokens, and API usage is billed by token count, not character or word count. Knowing your token count in advance helps you stay within a model's limits and estimate what a prompt or document will actually cost to process.
Tokenizers Can Differ
Different model families use different tokenizers, so the same text can produce a different token count depending on the model. Older and newer OpenAI models can use different encodings, and other providers use their own tokenizers entirely - which is why the cost estimator above clearly separates the tokenizer used for counting from the model used for pricing.