Token counter

What it does

A token counter tells you how much text you are really sending to a language model, because models bill and limit by tokens, not words. Paste a prompt, a document or a reply and the page counts tokens with o200k_base, the encoding behind GPT-4o through GPT-5, shows an estimate for Claude, and colours the text so you can see where each token starts and ends. The count runs in your browser; nothing is uploaded.

How to use it

1. Paste the text you plan to send, including any system prompt or instructions.
2. Read the OpenAI figure, exact for GPT-4o through GPT-5 and close for GPT-6, and the Claude estimate beside it.
3. Scroll the coloured preview to see how the text splits. Long words, numbers and code break into more pieces than you might expect.
4. Use Copy counts to paste the figures into a budget sheet or a ticket.

How to read the result

Suppose a support macro of 620 characters and 110 words counts as 142 OpenAI tokens. That is about 4.4 characters per token, typical for plain English. Sent 5,000 times a day on GPT-6 Luna at $0.10 per million input tokens, those 142 tokens cost about 7 cents a day. The same text estimated for Claude comes to around 178 tokens, because the estimate is deliberately conservative.

If the ratio drops to 2 or 3 characters per token, something in the text is expensive: JSON with lots of punctuation, URLs, base64 strings or text in a non-Latin script. Trim or summarise those parts first when you need to cut cost or fit inside a context window.

The number you need for planning is tokens per request multiplied by requests per month. Carry that into the AI API cost calculator to turn it into a bill.

Token budgets matter most when you are building a feature that calls a model on every user action. Our AI product features work starts with exactly this kind of measurement.