In AI language models, a token is a small chunk of text, often a short word, part of a longer word or a punctuation mark, that the model reads and writes. Models like GPT, Claude and Gemini measure everything in tokens: how much text fits in a conversation, how long an answer can be, and how much API usage costs. In English, 1,000 tokens is roughly 750 words.
What a token looks like
Tokenizers split text into the pieces that appear most often in training data. Common short words like “the” are a single token; longer or rarer words are split, so “tokenization” might become “token” and “ization”. Spaces and punctuation count too. Each model family has its own tokenizer, so the same sentence can be a slightly different number of tokens in GPT and in Claude.
Tokens per word in different languages
English is the most token-efficient language for most tokenizers because it dominates training data. Languages with long compound words, non-Latin scripts or rich grammar, such as German, Russian, Hindi or Japanese, often need noticeably more tokens for the same meaning. That means a non-English conversation fills the context window faster and costs more on the API.
Context windows: tokens as memory
A model’s context window is the maximum number of tokens it can consider at once, including your messages, its replies, system instructions and uploaded files. Flagship models in 2026 handle from about 128,000 to over a million tokens. When a chat or document exceeds the window, the app has to drop, summarize or search parts of it, which is one reason models sometimes “forget” earlier details or misread long PDFs.
Tokens and API pricing
Developers pay per million tokens, with separate prices for input (what you send) and output (what the model writes). Output is usually several times more expensive. Prices range from a few cents per million tokens for small or open models to tens of dollars for the most capable reasoning models. Reasoning models also generate hidden “thinking” tokens, which are billed as output.
A quick way to estimate cost: count the words, divide by 0.75 to get tokens, and multiply by the price per token for input and output separately.
How to use fewer tokens
- Send only the relevant part of a document, not the whole file.
- Start a new chat for a new topic instead of continuing a very long one.
- Ask for the length you need: “answer in 100 words”.
- Reuse a short system prompt instead of repeating long instructions.
- For high-volume tasks, use a smaller model that is good enough.
How subscription apps use tokens
ChatGPT Plus, Claude Pro and Google AI Pro do not bill per token, but their usage limits are still driven by token costs behind the scenes. Some all-in-one AI platforms show usage as tokens or credits, which makes the link visible.
Crypto “AI tokens” are something else
Searches for “AI tokens” also turn up cryptocurrency projects. Those are blockchain assets and have nothing to do with how language models process text.
To see where tokens fit in the bigger picture, read what a large language model is and how ChatGPT works.
Frequently asked questions
How many words is a token?
In English, one token is about four characters or three quarters of a word on average, so 1,000 tokens is roughly 750 words. Other languages often use more tokens per word.
Are AI tokens the same as crypto tokens?
No. In language models a token is a unit of text. Crypto tokens are digital assets on a blockchain and are unrelated, even when a project uses 'AI' in its name.
Do tokens matter if I use ChatGPT, not the API?
Yes, indirectly. Tokens decide how much text fits in one conversation and how long a file the model can read, even though subscription plans do not bill per token.