OpenAI Notes

IndexToken counter

Token counter

A token is the unit OpenAI bills. English lands near four characters per token. The box below is that estimate. The exact split is the official tokenizer.

Updated 2026-09-27

Characters
0
Words
0
Tokens, about
0

English at roughly four characters per token. Confirm the split on the official tokenizer before you budget a bill.

Open platform.openai.com/tokenizer when the bill matters. Paste the same text there. Different models do not share one vocabulary, so a count on an old encoder can miss a newer model by a little.

What the estimate is good for

  • Seeing that a pasted spec is thousands of tokens before you attach it
  • Comparing two prompts. The ratio is stable even when the absolute count is slightly off
  • Not for an invoice. The usage page on the platform is the charge

Input and output are both billed. A short prompt that asks for a long answer can cost more on the way out. Caps and prices are on the API pricing page, separate from how much ChatGPT Plus costs.

Questions

How many characters are in a token?

About four for English prose. Code, Chinese, and emoji are denser. Use the official tokenizer for a real count.

Does the ChatGPT app show a token count?

The API response and the platform usage page do. The consumer app shows plan limits, not a tokenizer readout on every message.

Related notes

  • OpenAI API and API keys

    Create an OpenAI API key at platform.openai.com. Usage is billed apart from ChatGPT Plus. new-api is a separate open-source gateway, not the official API.

  • OpenAI billing

    OpenAI billing splits in two: ChatGPT plan receipts in the app, and API spend on the platform organization overview.

  • GPT Realtime

    GPT Realtime is OpenAI’s low-latency speech model on the API. It is billed as API usage, not as a ChatGPT Plus add-on.

Sources