AI token counter

How to use it

1. Paste or type your text in the box, or open a .txt, .md, .csv or .json file from your device.
2. Pick the model you plan to use. Its current input price fills in automatically.
3. Read the token count and cost. They update as you type.

What a token is

AI models do not read letters or words. They read tokens: chunks of text that the model's tokenizer has learned are common. A short common word like "the" is one token. A longer word like "tokenization" may be split into two or three. Spaces and punctuation are usually part of a token too.

Providers bill by the token, and every model has a limit on how many tokens it can handle at once. That makes the token count the single most useful number to know about a prompt, a document or a chat history before you send it.

How this counter works

The count shown first is exact for OpenAI models that use the o200k_base encoding, which covers GPT-4o, GPT-4.1, the GPT-5 family and the o-series reasoning models. The tokenizer is the same open-source algorithm OpenAI publishes, running entirely in your browser. Your text is never uploaded.

Anthropic (Claude) and Google (Gemini) use their own tokenizers and do not publish them for offline use. For those models the number shown is an estimate based on the GPT count. In everyday English text the difference is usually within 10 to 20 percent, but it can be larger for other languages, code and unusual formatting. If you need an exact figure for Claude or Gemini, both providers offer a free token-counting endpoint in their APIs.

Worked example

A typical one-page business email of about 350 words comes to roughly 450 to 500 GPT tokens. Sending it to a model priced at $2.50 per million input tokens costs about $0.0012. That is tiny for one email, but a customer-support bot that sends a 3,000-token knowledge base with each of 50,000 monthly questions spends 150 million input tokens a month, about $375 at the same price. The counter helps you spot that kind of cost before it lands on an invoice.

Why the same text gives different counts

• Language. Tokenizers are trained mostly on English. Malay, Tamil, Chinese and Thai text often needs 1.5 to 3 times as many tokens per word.
• Numbers and code. Long numbers, IDs, URLs and source code split into many small tokens.
• Formatting. Markdown symbols, tables and repeated whitespace all count.
• Hidden overhead. Chat APIs add a few tokens per message for roles and formatting, and tool definitions count as input. Treat the figure here as the content cost, then add a small margin.

Ways to use fewer tokens

Remove boilerplate that the model does not need, such as repeated disclaimers or long signatures. Summarise old chat turns instead of resending them in full. Ask for a specific output length. If the same long instructions go out with every request, check whether your provider offers prompt caching: cached input is billed at a fraction of the normal rate, and you can see those rates on our model price comparison.

Questions