Context Window Budget Planner
Token counters give you a number. This tells you whether your document actually fits before the model starts truncating it. Paste your text or enter a word count and pick a context window size. Nothing uploaded.
Results
Context Window Usage
-
A Note On Accuracy
This tool estimates tokens using a rough ratio of 1.33 tokens per word. Actual tokenization varies by model, language, and content type (code and non-English text often tokenize less efficiently than plain English prose). Use this as a planning estimate, not an exact count.
Learn more: tokens, context windows, and what this estimate can tell you
Where the 1.33 figure comes from
Models read text as tokens rather than words, and a token is usually a fragment of a word. This planner multiplies your word count by 1.33 and rounds to the nearest whole token, so 1,000 words reads as 1,330 tokens and 7,500 words reads as 9,975.
That ratio is a rule of thumb for ordinary English prose, not a property of any particular model. Two models given the same paragraph will often return different token counts, so use the number here for planning rather than as an exact preflight check before an API call.
Where the word count stops being meaningful
The "Paste text" tab counts words by splitting on whitespace. That works for prose and badly misreads code, where punctuation and dotted method chains run together without spaces. A single line of JavaScript can register as two or three words while tokenizing into a dozen or more pieces, which pushes the estimate far too low.
Text written without spaces between words has the same problem, only worse. A full paragraph of Japanese or Chinese can count as one word, so the token figure comes out close to meaningless. For either case, switch to the "Word count" tab and enter a number from a real tokenizer for the model you are targeting.
Room left has to cover the reply too
"Room left" subtracts your estimated tokens from the context size you picked. That remainder is not spare capacity for more input: it also has to hold your system prompt, any other documents you attach, and the model's own answer. A document sitting at 90% of the window leaves very little space for a long response.
Go past the limit and the bar turns red, "Room left" switches to "Over by" plus the overshoot, and the caption tells you how many tokens you need to cut.
Fitting is not the same as being read evenly
Whether your document fits is the first question, not the last one. Nelson Liu and colleagues found in Lost in the Middle: How Language Models Use Long Contexts that performance is highest when the relevant information sits at the start or the end of the input, and degrades significantly when the model has to retrieve it from the middle.
They saw the same pattern in models built specifically for long contexts. So a document that lands at 40% of the window has not left you 60% of free quality to spend. Moving your instructions and the passages that matter toward the edges tends to help more than adding material.
FAQ
Why do I have to pick a context size instead of picking a model?
Context limits change between model families and between versions of the same family, so a hardcoded model list here would go stale within months. The dropdown covers the common sizes from 8,000 up to 2,000,000 tokens, and "Custom" takes any number you type.
What is "Copies that fit" actually for?
It divides the context size by one document's estimated tokens and rounds down, so a 10,000-token document in a 128,000-token window shows 12. It assumes nothing else is in the window, so read it as an upper bound when you are working out how many reference files you could paste into one prompt.
Should I try to get the token count as low as possible?
Not at the cost of being clear. Trimming pays off when you are near the limit or billed per token, but cutting context the model actually needs costs you more than the tokens save you. Given the position effect above, moving important text to the start or the end is usually worth more than shaving a few hundred tokens out of the middle.