Key points
- A token is a chunk of text — a word, part of a word, a space or punctuation — that an AI model reads and writes.
- Rule of thumb for English: 1 token ≈ 4 characters ≈ 0.75 words (OpenAI); 100 tokens ≈ 60–80 words (Google).
- Token counts differ between models because each family uses its own tokenizer.
- API prices and context windows are measured in tokens, and output tokens usually cost several times more than input tokens.
What is a token?
AI language models don’t process text letter by letter or word by word. Before a model sees your message, a component called a tokenizer splits it into pieces drawn from a fixed vocabulary. Common words like “the” or “house” are often a single token. Longer or rarer words are split into several: an unusual name or a technical term might become three or four tokens. Punctuation marks are tokens, and a space is usually attached to the start of the following word.
Each token is mapped to a number, and it’s these numbers the model actually works with. When the model replies, it generates one token at a time, which the app turns back into readable text. That’s why we say models “read and write tokens”. If you want the bigger picture of how this fits into the way a model works, start with What is an LLM?
How many words is a token?
It depends on the language and the kind of text, but providers publish rules of thumb for English:
- OpenAI says one token is approximately four characters or 0.75 words of English text.
- Google says that for Gemini models a token is about four characters, and 100 tokens equal about 60–80 English words.
- Anthropic doesn’t publish a fixed ratio for Claude. Its documentation recommends counting with its token-counting API and notes that its newer models produce roughly 30% more tokens for the same input than its earlier models.
So a 750-word article is roughly 1,000 tokens by OpenAI’s rule, and somewhere between about 940 and 1,250 tokens by Google’s range. For quick estimates, our free token counter applies these published rules to any text you paste.
Why token counts differ
Tokenizers are built from data. The pieces that appear most often in the training text become single tokens, so whatever is common in the data gets encoded efficiently. Several consequences follow:
- Different models, different counts. OpenAI, Anthropic and Google use different tokenizers, and they update them between model generations.
- Languages vary. English is usually encoded efficiently. Many other languages need more tokens for the same meaning, which means they also use more of the context window and cost more through an API.
- Code, numbers and symbols are often split into many small tokens. A long number may be several tokens; unusual formatting adds more.
- Whitespace and formatting count. Large tables or heavily indented code use more tokens than the same information as plain sentences.
Why tokens matter to you
1. Context windows
Every model has a maximum number of tokens it can consider at once — its context window. That limit covers your messages, the chatbot’s replies, any documents you upload and hidden system instructions. Many current flagship models list context windows of one million tokens through their APIs, but consumer apps may apply their own limits. When a conversation exceeds the window, older parts are dropped or summarised, and the model may “forget” details from early in the chat.
2. Answer length
Models also have a maximum number of output tokens per reply. If an answer stops mid-sentence, it may have hit that limit — ask it to continue.
3. API costs
If you build with AI through an API, you pay per token, with separate prices for input (what you send) and output (what the model generates). Prices are quoted per million tokens. On OpenAI’s pricing page, for example, GPT-6.2 Sol is listed at $2.00 per million input tokens and $10.00 per million output tokens; on Anthropic’s pricing page, Claude Sonnet 5.5 is listed at $2 input and $10 output per million tokens (both checked November 19, 2026). Output costing about five times as much as input is common, so long answers are the main cost driver. Use our API cost calculator to compare models for your workload.
4. Usage limits in chat apps
Free and paid chatbot plans have usage limits. Providers don’t always publish the exact formula, but longer conversations and large uploads generally use more of your allowance than short questions, because there are more tokens to process.
Worked example
Suppose you upload a 6,000-word report and ask for a 500-word summary. By OpenAI’s rule of thumb, the report is about 8,000 tokens and the summary about 670 tokens. If you then ask five follow-up questions in the same chat, the app may re-send the whole conversation, including the report, each time — so the total tokens processed can be many times the size of the report. In a chat app that affects how quickly you reach usage limits; through an API, it affects your bill.
Tips for using tokens wisely
- Paste only what’s relevant. If you need help with one section of a document, paste that section.
- Start fresh chats for new topics. Long conversations carry a growing token load and can degrade answer quality.
- Ask for concise answers when you don’t need detail. “In under 150 words” saves output tokens.
- For APIs, use caching and batch modes where providers offer them; Anthropic, OpenAI and Google all list discounted batch and cached-input rates on their pricing pages.
- Count, don’t guess, for billing. Rules of thumb are fine for planning; use the provider’s tokenizer or counting endpoint when money depends on it.
Want to see it in action? Paste a paragraph into our token counter, then paste the same paragraph translated into another language and compare.
Remove this box by adding your content here, or hide all notes with
"showEditorNotes": false in site.config.json.Summary
Tokens are the currency of AI language models. You don’t need to count them for casual chatting, but knowing the rough conversion — about three-quarters of a word per token in English — helps you understand context limits, why long chats get slow or forgetful, and how API pricing works.
Frequently asked questions
How many words is a token?
For English text, OpenAI’s rule of thumb is about 0.75 words (or about 4 characters) per token, so 1,000 tokens is roughly 750 words. Google says 100 tokens is about 60–80 English words for Gemini.
Are tokens the same in every AI model?
No. Each model family has its own tokenizer, so the same text can produce different token counts. Anthropic notes that its newer models produce roughly 30% more tokens for the same text than its earlier models.
Do spaces and punctuation count as tokens?
Yes, they’re part of the text the model processes. Often a space is merged with the following word into one token.
Why do other languages use more tokens?
Tokenizers are built from training data that is often dominated by English, so English words are usually stored efficiently. Many other languages, especially non-Latin scripts, are split into more pieces.
Do tokens matter if I use the free ChatGPT or Claude app?
Indirectly. You don’t pay per token, but usage limits and the amount of text a chatbot can handle in one conversation are based on tokens.
Sources
- OpenAI — Key concepts (tokens) (opens in a new tab)
- Google — Understand and count tokens (opens in a new tab)
- Anthropic — Token counting (opens in a new tab)
- OpenAI API pricing (opens in a new tab)
- Anthropic — Claude API pricing (opens in a new tab)
Facts, prices and features were checked against these sources on November 19, 2026. AI products change quickly; if you spot something out of date, please tell us.