AI Token Counter
Estimate how many tokens a text uses in OpenAI GPT, Anthropic Claude, Google Gemini and Meta Llama models from its word or character count, using each provider’s published rule of thumb.
Last updated: October 2026
Estimated tokens
1,334 tokens
Formula below · 4 sources (help.openai.com, platform.claude.com, ai.google.dev, arxiv.org) · Updated Oct 2026
Compare with similar
About this calculator
Language models read text as tokens: whole words, pieces of words, punctuation and spaces. Every model family has its own tokenizer, so the same text is a different number of tokens in each. This calculator estimates the count from the length of the text with the rules of thumb the providers publish for English:
- OpenAI GPT: about 4 characters or ¾ of a word per token (100 tokens ≈ 75 words). - Anthropic Claude (the tokenizer introduced with Claude Opus 4.7 and used by current models): 1 million tokens ≈ 555,000 words or 2.5 million characters, so about 0.555 words or 2.5 characters per token. Anthropic notes this tokenizer produces about 30% more tokens than earlier Claude models. - Google Gemini: about 4 characters per token; 100 tokens ≈ 60–80 English words (the calculator uses the midpoint, 70). - Meta Llama 3: 3.94 characters per token on English text (Llama 3 paper). Meta gives no words figure; because Llama 3’s vocabulary is built on 100,000 tokens from OpenAI’s tiktoken, words are converted at OpenAI’s 5.33 characters per word (4 ÷ 0.75).
tokens = words ÷ words per token, or characters ÷ characters per token, rounded up.
Code, JSON, markup and most languages other than English break into more tokens per word than English prose. No provider publishes a single ratio for them, so the content-type options add a planning margin (+25% for code, +50% for other languages); these margins are the calculator’s assumption, not a measured constant. Languages written in Chinese, Japanese or Korean characters can use one token or more per character.
These are estimates. Exact counts need the model’s own tokenizer: OpenAI publishes its tokenizers (tiktoken and the online Tokenizer tool); Claude’s tokenizer is not public, but Anthropic’s token-counting API returns the exact count for a message on a given model; Gemini has a count_tokens method.
How to use
Enter the length of your text, say whether it is in words or characters (a word processor shows both), pick the content type and the model family for the headline number. The tiles show the same text in all four families. Example (the defaults): 1,000 words of English prose. OpenAI GPT: 1,000 ÷ 0.75 = 1,333.3, so about 1,334 tokens. Claude: 1,000 ÷ 0.555 = 1,801.8 → 1,802 tokens. Gemini: 1,000 ÷ 0.7 = 1,428.6 → 1,429 tokens. Llama 3: 1,000 ÷ 0.73875 = 1,353.6 → 1,354 tokens. Claude counts about 1.35× as many tokens as GPT for the same words. Example 2: 10,000 characters for Claude: 10,000 ÷ 2.5 = 4,000 tokens, against 10,000 ÷ 4 = 2,500 for GPT (1.6×). Example 3: 1,000 words of code for GPT: 1,333.3 × 1.25 = 1,666.7 → 1,667 tokens.
Frequently asked questions
How many tokens is 1,000 words?
About 1,333 tokens for OpenAI GPT models (¾ of a word per token), about 1,430 for Gemini (60–80 words per 100 tokens), about 1,350 for Llama 3 and about 1,800 for current Claude models (555,000 words per million tokens). Code and non-English text use more.
Does Claude use more tokens than ChatGPT for the same text?
Yes, by these published ratios. Anthropic says its current tokenizer fits about 555,000 English words in 1 million tokens, against roughly 750,000 words for OpenAI’s ¾-word rule, so the same English text is about 1.35× as many Claude tokens. Per-token prices are not the whole comparison: compare the cost of the same text, as the ChatGPT vs Claude cost calculator does.
How do I count Claude tokens exactly?
Claude’s tokenizer is not published, so no offline library gives the exact number. Anthropic’s token-counting API (the count_tokens endpoint) returns the input token count for a message on a specific model, free of charge within rate limits, and every API response reports the input and output tokens it billed.
Are tokens the same as words?
No. A common English word is often one token, but long or rare words split into several, and spaces, punctuation and numbers take tokens too. OpenAI’s help centre notes that a token count is not the same as a word count and that the same text gives different counts on different models and languages.
Related calculators
Sources & references
Think a result is wrong? Report a problem with this calculator.