Context window visualizer
Turn a token limit into things you can picture: pages, novels, hours of meeting transcript and lines of code. Then compare the context windows of current models on one scale.
Pick a size
0.75 suits English prose. Use about 0.5 for code-heavy or non-English text.
This much context holds about
96,000 words
Compared with familiar texts
Context windows of current models
Log scale: each step to the right is about ten times more text. The marker shows your size.
How the conversions work
The visualizer converts tokens to words with your words-per-token ratio, then uses fixed rules of thumb: 500 words to a printed page, 90,000 words to an average novel, about 9,000 spoken words to an hour of meeting transcript (150 words a minute), and about 10 tokens to a line of code. All of these vary, so read the numbers as orders of magnitude.
The familiar-text comparisons use commonly cited word counts: about 77,000 words for the first Harry Potter book, about 587,000 for an English translation of War and Peace and about 884,000 for the complete works of Shakespeare.
What it means in practice
- Room for the answer. The response usually shares the window, so a prompt that fills it leaves nothing for output.
- Price per call. Every token you send is billed every time. A 500,000-token prompt on a model at $2 per million input tokens costs $1 per call before any output.
- Long-prompt surcharges. Several providers charge higher rates above a threshold such as 200K or 272K tokens; check the model page.
- Recall. Test whether the model finds facts deep in your documents before you rely on it.
More in context windows explained. To measure a real document, use the token counter.
Need hands-on help building it?
RAG, agents, fine-tuning or a production chatbot: describe the project and we'll suggest an approach and, if you want, partners who build it.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.
Frequently asked questions
How many words is 128K tokens?
About 96,000 words of English, using the rule of thumb of 0.75 words per token. That's roughly 190 pages at 500 words per page, or slightly more than an average novel. Code and non-English text fit less.
Does the context window include the model's answer?
Usually yes: the prompt and the response share the window, and many models also cap the response separately with a maximum output limit. Leave room for the answer, and for reasoning tokens if the model thinks before replying.
Is a bigger context window always better?
No. Long prompts cost more on every call, take longer to process, and models don't always use information buried in the middle of a long prompt reliably. Sending only the relevant parts is often cheaper and more accurate.