AI model context window calculator

Every language model has a context window: the maximum number of tokens it can handle at once. Everything goes in it: your instructions, the documents you attach, the conversation history, tools and also the model’s answer (with its reasoning, if it does any). Go over and the request fails or the model cuts off.

Enter how much you send (in tokens, words, characters or pages) and how much answer you expect, pick a model and you see how much of its window you use, how much is left and in how many models on the list it fits. The data carries a review date. Everything runs in your browser.

What you send is converted to tokens (if you do not give tokens, using the approximate English equivalence: 1 token ≈ 0.75 words ≈ 4 characters), the expected answer is added and the total is divided by the model’s window. If the total is below the window, it fits; the difference is the free tokens, which are turned into approximate words and pages.

Each model’s data (window and maximum output) lives in a file with its review date and official sources.

What the context window is

It is the model’s “working memory”: the total tokens it can consider when generating an answer. It includes everything in the request (system instructions, messages, documents, tool results, images) and also what the model generates, reasoning included. It is not the same as the data it was trained on.

Context window versus maximum output

They are two different limits: the window limits the sum of input and output; the maximum output limits how much the model can write in a single answer. A model with a 1-million window may have, say, a 128,000-token maximum output.

More context is not always better

As the token count grows, the model’s accuracy and recall degrade: Anthropic calls this “context rot”. Send what is needed and do not fill the window for its own sake; techniques such as summarising, trimming old results or retrieving only the relevant fragments help.

How to estimate tokens

The most reliable way is the provider’s token counter. As a rough guide, in English a token is about 0.75 words (about 4 characters); in other languages there are more tokens per word. Each provider tokenizes differently: the same text gives different figures depending on the model, and Claude 4.7 and later produce about 30% more tokens than earlier ones for the same text.

Common mistakes

• Forgetting that the answer and reasoning also take up the window.
• Trusting the words-to-tokens conversion in languages other than English.
• Filling the window to the brim: leave a margin.
• Comparing windows without comparing maximum output.
• Treating limits as fixed: they change with each model.

Limits of this calculator

• The data are those of the standard API at the review date; they do not include extended contexts in beta or by contract, nor the limits of chat apps or usage tiers.
• There is no maximum-output figure for Claude Haiku 4.5 in the source consulted.
• The OpenAI and Google figures were taken from their pages with a tool that summarises the content: check them on the official page.
• The conversion of words, characters and pages to tokens is approximate and based on English.
• It does not count images, audio or tools, which also use tokens.

Notice

This is an estimate. Count tokens with the provider’s tool before fitting a case right to the limit.