AI & Dev
Token Counter
Paste text and see how many tokens it costs across GPT, Claude, and Gemini — plus what it does to your API bill and context window.
How it works
Behind the numbers
Language models don't read words — they read tokens, chunks of text the model was trained on. Tokenization is invisible, but it decides two things you pay for: how much of the context window your text fills, and your API bill, since providers charge per thousand tokens.
This counter estimates tokens with a heuristic tuned per model family (roughly 4 characters per token for English on GPT-style models, adjusted for Claude and Gemini). It is an estimate, not the exact tokenizer — accurate enough for budgeting prompts and staying inside context limits, usually within a few percent.
Switch models to compare tokenizers, watch the context-window bar to see how much of the model's memory your text consumes, and check the cost estimate to price a prompt before you run it a thousand times in production.
FAQ
Questions, answered
What is a token in AI?
A token is a chunk of text — on average about three-quarters of an English word. Common words are often one token; rare words can split into several. 'Tokenization' is how models slice text before processing it.
Why does token count matter?
Two reasons: every model has a maximum context window measured in tokens, and API providers bill per thousand tokens in and out. Long prompts cost real money at scale.
Is this as accurate as tiktoken?
No. This is a fast heuristic estimate, typically within a few percent. For exact counts — for example, to guarantee you fit a context window — use the model's official tokenizer.
Do different models tokenize text differently?
Yes. Each model family trains its own tokenizer vocabulary, so the same text produces different token counts on GPT, Claude, and Gemini. That is why the model switcher exists.
How can I reduce the token count of a prompt?
Write concisely, cut repetition, prefer short instructions over examples where possible, and strip formatting the model doesn't need. Measure before and after with this counter.