Token counter. GPT, Claude, Gemini and more.

Paste text to count its tokens with OpenAI's tokenizer, see the estimated count on Claude and other models, and what sending it costs on each. It all runs in your browser.

Counted in your browser. Your text isn't sent anywhere.

OpenAI tokens
…
o200k_base, exact
Claude tokens
…
Opus 4.7 and later, estimate
Characters
0
including spaces
Words
0
split on spaces

What this text costs per request

Each model's own token count, cheapest first. Struck through: the lab's list price.

Live prices in US dollars per request. Counts for labs other than OpenAI are scaled from OpenAI's count by the ratios we measure on real traffic, so they're estimates; each API response's usage field has the exact figure. For a whole month of requests, use the LLM cost calculator.

How it works

What a token is. And why labs count differently.

Models read text as tokens: common words are one token, rarer words split into pieces, and spaces and punctuation ride along. In English a token averages about four characters, or three quarters of a word. Code, numbers and most other languages need more tokens for the same length.

Each lab has its own tokenizer. OpenAI publishes its tokenizers, so GPT counts here are exact for o200k_base. Claude's tokenizer isn't public, and Claude Opus 4.7 and later count about 1.53 tokens for every OpenAI token on our traffic, and we count Gemini at about 1.1 times OpenAI's figure. The same prompt can come to a different number of tokens on each model, which is why the list above counts per model.

Every token is billed. Input tokens are everything you send, including the system prompt, tool definitions and the conversation so far; output tokens are what comes back, thinking included, at several times the price. To cut the bill, see how prompt caching works or seven ways to spend less.

FAQ

Token questions.

Anything else? Ask our team and we’ll answer by email.

How many tokens is 1,000 words?

In ordinary English, about 1,300 tokens with OpenAI's tokenizer, since a token averages about three quarters of a word. Claude's newer models count about 2,000 for the same text. Code, numbers and most other languages take more tokens per word.

Which tokenizer does this use?

o200k_base, the newest tokenizer OpenAI has published in its tiktoken library, used by GPT-4o and later models. OpenAI hasn't published a separate tokenizer for its newest models, so treat their counts as close; the usage field of a real response is always the exact number.

How do I count Claude tokens exactly?

Send the request to Anthropic's token counting endpoint, POST /v1/messages/count_tokens, which returns the input tokens the request would use. The figure here is an estimate, scaled from OpenAI's count by the ratio we measure on real Claude traffic.

Why do Claude's newer models count more tokens?

Claude Opus 4.7 and later models, including the 5.x models, use a newer tokenizer. On our traffic they count about 1.53 tokens for each token OpenAI's tokenizer finds in the same text, so compare per-token prices with that in mind.

Is my text sent anywhere?

No. The tokenizer is downloaded to your browser and the counting happens there. Nothing you type or paste is sent to Zurelay or anyone else.

Do images and files count as tokens?

Yes. Images, PDFs and audio are converted to tokens too, by rules that differ for each lab, and they aren't counted here. The usage field of each API response includes them.

More pricing

Stop paying list price.
Start saving today.

Create a key in seconds, point the SDK you already use at Zurelay, and every request costs up to 90% less from the first token.

Get your API keyTalk to an engineer