Free · Live · Nothing uploaded

AI token counter

Count tokens for GPT-4o, GPT-5, GPT-3.5/4, Claude and Gemini live as you type, with context-window fit and an API cost estimate. Real GPT tokenizer, runs entirely in your browser.

0tokens
0characters
0words
0%context used
$0.00est. cost
Exact
Context window & cost settings

Defaults are rough July 2026 examples for a representative model in each family — prices and context limits change often, so check your provider for current numbers.

🔒 Your text is counted entirely on your device — never uploaded, even for the exact GPT tokenizer.

How to count AI tokens

Check how many tokens a prompt costs before you send it — for context-window limits, API cost, or fine-tuning data prep.

🧩

1. Pick a model

Choose GPT, Claude, Gemini or another model family — each tokenizes text differently.

⌨️

2. Paste your prompt

Token, character and word counts update live with every keystroke.

📐

3. Check fit & cost

See what share of the context window you're using and an estimated API cost.

Useful for trimming a prompt to fit a model's context window, sanity-checking API costs before a big batch job, or estimating fine-tuning data size. GPT-4o, GPT-5, GPT-3.5 and GPT-4 counts use the same byte-pair-encoding tokenizer those models use internally; Claude, Gemini, Llama and Mistral counts are estimates, since those providers don't publish a downloadable tokenizer. Everything runs in your browser, so even confidential prompts and proprietary code stay private. For plain word and character counts, try the Word Counter.

Token counter FAQ

What is a token?

A chunk of text a model reads or writes — often a word, part of a word, or punctuation. English averages about 4 characters, or ¾ of a word, per token.

Are the GPT counts exact?

Yes — GPT-4o/GPT-5 and GPT-3.5/GPT-4 counts use the real tokenizer those models use internally, run entirely in your browser.

Why are Claude and Gemini counts estimates?

Anthropic and Google haven't published a downloadable tokenizer, so those are estimated from their documented characters-per-token ratios — usually within a few percent.

Is my text uploaded?

No — counting happens entirely on your device, including the real GPT tokenizer.