Free AI Context Window Checker
Will it fit? A yes or no, and what comes after
Every AI model can only consider a limited amount of text at once, called its context window. Your instructions, any documents you paste, the conversation so far and the answer it writes all have to fit inside. This checker estimates the tokens in your text, or takes a token count you already have, reserves room for the reply, and shows which popular Claude, GPT and Gemini models it fits. It runs in your browser, so your document stays private.
How to use it
- Paste your text, or type a token count into the second box.
- Set how many tokens to leave for the reply. A short answer needs a few hundred, a long report or a block of code can need thousands.
- Read the table. Each model shows its window, how much of it your text uses and whether it fits.
What a window means in real documents
| Window | Roughly fits, in English |
|---|---|
| 128,000 tokens | About 96,000 words, around 190 pages |
| 200,000 tokens | About 150,000 words, around 300 pages |
| 1,000,000 tokens | About 750,000 words, several long books |
Code, tables and non-English text use more tokens, so fewer pages fit than these figures suggest.
Fitting is not the same as working well
A model that accepts a huge document is not guaranteed to use every part of it equally well. Details buried in the middle of a very long input can be missed, answers get slower and every request costs more. If you can, send only the relevant sections, put your question at the end of a long document, and ask the model to quote the part it relied on, so you can check it. A shorter, focused input is usually more reliable than a maximum-size one.
When your text is too big
- Split it. Process the document in sections and combine the results.
- Summarise first. Condense long background into a short brief, then work from that.
- Use search over documents. Systems that retrieve only the relevant passages, rather than loading everything, scale far beyond any window.
- Choose a larger model if your provider offers one, remembering the higher cost.
Long chats fill the window quietly
In a chat, every earlier message stays in the window, so a long conversation slowly uses up space. When answers start to ignore earlier instructions, the window may be nearly full. The remedy is to start a new conversation with a short summary of what matters. This is also cheaper, since fewer tokens are sent each turn.
The reply allowance, and how to set it
The box for leaving room for the reply starts at 4,000 tokens, which suits a typical answer of a few paragraphs. Raise it if you expect long output, such as a full article, a long piece of code or a translated document, since the reply can be as large as the text you send. Lower it for short answers, such as a yes or no and a sentence of reasoning. The models are listed from the largest window to the smallest, each with a bar and a label that reads Fits with a percentage used, or Too large with the share of the window you would need.
Three quick scenarios
- A 10-page report, about 7,000 tokens. Fits comfortably in almost every model, using a small share of a large window.
- A 300-page book, about 120,000 tokens. Fits only in models with a large window, and leaves little room for a long reply in a 128,000 token one.
- A whole codebase, 500,000 tokens or more. Fits only in the very largest windows, and even there it is often better to send only the relevant files.
About the list of models
The models and their windows are maintained by hand, and the date they were last checked is shown in the tool. Providers update models often, and some offer larger windows only on certain plans or through the API, so confirm the limit for your own account. For exact counts use the GPT token counter, and for the money side, the AI model price comparison.
FAQ
The maximum number of tokens an AI model can consider at once, including input and output.
The API rejects the request, or a chat app trims or summarises older content. Split the text or choose a model with a larger window.
Yes. That is why the tool reserves room for it.
It is an estimate for most models. Enter an exact count from a tokenizer if you have one.
No. The check runs in your browser.
The reply uses part of the same window as your prompt, so a bigger reply allowance means your input has less room, and some models may stop fitting.