Token counter
Free token counter for GPT, Claude, and Gemini. Paste any text to see roughly how many tokens it costs, alongside characters, words, and characters per token. Counts are estimated in your browser — nothing you paste is uploaded.
How to use token counter
- 01Paste your text
Drop in a prompt, document, code file, or chat transcript. Counts update as you type.
- 02Read the three token counts
GPT, Claude, and Gemini tokenize differently, so the same text costs a different number of tokens on each.
- 03Check characters per token
English prose sits near 4. Code and JSON drop lower because punctuation splits into its own tokens.
- 04Pick a model
The model you select drives the context-window percentage and the indicative input cost below it.
- 05Verify before you budget
For billing-critical numbers, confirm against the provider's own counter — the links sit under the results.
Examples
Formula
Text is split into runs and each run is priced separately: a word up to five letters costs one token and adds another every three letters beyond that, digits group in threes, punctuation runs merge in pairs, and CJK costs about 0.7 tokens per character. The total is then scaled by a per-provider ratio. Two rules of thumb approximate the same thing for English prose: about 4 characters per token, or about 1.33 tokens per word.
tokens ≈ Σ(run cost) × provider ratioFrequently asked questions
How accurate is this token counter?
It is an estimate, not an exact count. On English prose it typically lands within 10% of the real number, within 15% on code and JSON, and within 20% on Chinese, Japanese, and Korean. That is close enough to size a prompt or sanity-check a bill, but if a number is billing-critical, verify it against the provider's own counter.
How many tokens is one word?
For English prose, roughly 1.33 tokens per word, or about 4 characters per token. Short common words are usually a single token, while long or unusual words split into several pieces. Punctuation, line breaks, and indentation all cost tokens too, which is why code and JSON run much denser than ordinary writing.
Why do GPT, Claude, and Gemini show different counts?
Each provider trains its own tokenizer on a different vocabulary, so the same sentence splits differently on each. The gap is usually within about 10%, but it widens on code, non-English text, and unusual formatting. Providers also change tokenizers between model generations, so counts can shift when you upgrade.
Does the text I paste get uploaded?
No. The whole calculation runs in your browser using a local character model, and nothing is sent to a server or to any model provider. That is also why the count is an estimate rather than exact: producing an exact count would mean either shipping a large tokenizer vocabulary or calling a provider API with your text.
Why does the cost figure ignore the model's response?
Pasted text only tells us the input side. Output length depends on what you ask for and how the model answers, and output tokens usually cost several times more than input tokens. Rather than invent a response length, the tool shows input cost only. Treat every price here as indicative — rates change often.
Last updated . Built by Tooligan.