Token counter
Models bill by the token, and a token is not a word — the same paragraph is a different number for different models. Paste anything below to count it exactly and get a cost for both halves of the call: what you send, and what the model writes back. Current rates for GPT-6, Claude Fable and Opus 5.5, and Gemini 4 come prefilled, so there is nothing to look up, and every one of them is yours to overwrite. Nothing is sent anywhere; the counting happens in your browser.
Your text
Nothing leaves your browser. The counting happens here, on your machine.
Models split text differently, so the same words are a different number of tokens depending on where you send them. The first count downloads that tokenizer (about 1 MB), once per visit.
What that costs
Per run
—
- Tokens
- 0
- Characters
- 0
- Words
- 0
Paste something above to count it.
Rate, per million tokens
Checked 14 November 2026
Filled in for you, and yours to change — an edit is kept for next time. If your bill disagrees with this, your bill is right: check the provider’s own pricing page and type what it says.
What this misses. Tiered and regional pricing, batch and cached-input discounts, images and audio, and the reasoning tokens some models bill for but never show you. It is a good estimate of a text call, not an invoice.
Anthropic and Google do not publish their tokenizers as data anyone can run, and OpenAI has not said which one GPT-6 uses, so counts for Claude, Gemini and GPT-6 are produced with the GPT-5.6 tokenizer and marked with a ≈. They land in the same neighbourhood, but if you need the exact figure, use the provider’s own token-counting endpoint.
Read next
Get the ones worth knowing about
An occasional email about tools that turned out to be worth the switch — and the ones that did not. No launch roundups.
Your address goes to Buttondown, who send the emails on our behalf. One click unsubscribes, and the list is never sold or shared.