Quick Answer
A context window is the total amount of text a model can work with in one request. Claude measures it in tokens, which are hard to picture. Converting to words, pages, and books makes it concrete. Because Claude's tokenizer is not public, the figures below are ranges based on our measured English prose average and the usual gap between OpenAI and Claude token counts.
Claude Context Windows in Words and Pages
| Window | Models | Words (estimate) | Single-spaced pages |
|---|---|---|---|
| 200K tokens | Claude Haiku 4.5 | 110,000 to 129,000 | 220 to 260 |
| 1M tokens | Claude Opus 5, Claude Sonnet 5, Claude Fable 5.1 | 550,000 to 645,000 | 1,100 to 1,290 |
What Fits in 1M Tokens
- Novels: with a typical novel at 70,000 to 100,000 words (see how many words are in a novel), 1M tokens holds about 5 to 9 books.
- Theses: a PhD thesis of 80,000 to 100,000 words fits several times over.
- Reports and contracts: hundreds of pages of documents can be read in a single request.
- Codebases: code uses more tokens per line than prose, so the same window holds less than the word figures suggest.
What Uses Up the Window
The window is shared by everything in the conversation, not only the document you paste. System instructions, earlier turns, tool results, and Claude's own reply all count. Long chats fill it gradually, which is why a document that fits in a fresh conversation may not fit late in a long one. Replies also have a separate maximum length, so a full window does not mean an equally long answer.
Other languages and number-heavy text use more tokens per word, so fewer words fit. In our measurements, Turkish text needed 42% more tokens than English on the same topics. To see how a specific document measures up, paste it into the Token Counter, which shows its fill for both 200K and 1M windows.