Skip to content

Context fit

A context window is the whole of what a model can hold at once — your document, the conversation and its own reply. Drop a file or paste text to see how much of each model's window it would take.

Your file never leaves this page. There is no server runtime behind this site: the PDF is parsed and the tokens are counted by your own browser. Nothing is uploaded, nothing is logged, nothing is stored — you can read the network panel and check.

drop a file here — .txt, .md or a PDF

Result

Nothing measured yet. The encoding dictionary is only downloaded once there is text to count, and the PDF engine only when you hand us a PDF.

How the check works

A context window is counted in tokens, and every model has its own size — from 128,000 to about a million. This tool counts your text the way each model would: with the real published dictionary where one exists, and with a clearly marked vendor-derived estimate where it does not. It then shows the share of each window your text would occupy. The share is deliberately conservative in two ways: estimates are chosen to over-count, and the window also has to hold the conversation and the reply — which is why the page warns that a document at 90% of the window is not usable the way one at 20% is.

Common questions

1 Does my file leave my browser?
No. The PDF is parsed and the tokens are counted by your own browser; the site is static files with no server runtime, so there is nowhere to upload to. You can watch the network panel while you use it.
2 Which files can I drop?
.txt, .md and PDF files with a text layer. A scanned PDF without a text layer yields no text — the tool detects that and says so instead of reporting “0 tokens, fits everywhere”, which would be the wrong answer.
3 Why are some rows marked as estimates?
Anthropic, Google and OpenAI (for its newest models) have not published their tokenizers. Those rows divide the character count by the vendor’s own characters-per-token figure, chosen to over-count rather than under-count — so a document the tool says fits, fits.
4 If a document fits, can I actually use the whole window?
Not quite. The window also has to hold the system prompt, the conversation so far and the model’s own reply. A document at 90% of the window is not usable the way one at 20% is — the share column is a budget, not an invitation to fill it.

Want to see the individual tokens rather than the totals? Token counter shows the split.