Claude's Context Window: What a Million Tokens Actually Changes in Your Work
A plain-language guide to what a context window is, how much text Claude holds at once, and why it matters for long contracts, reports, and ongoing projects.
You paste a long contract into the chat, ask for a summary, and the model answers as if it only read the first page. Or an hour into a conversation it forgets what you agreed on at the start. That is not the model being dumb. It is the limit of how much it can hold in a single conversation. That limit has a size, and it is called the context window.
What is a context window, in plain terms
The context window is the amount of text the model keeps in front of it right now: your question, the files you attached, the full history of this chat, and its own replies. All of it has to fit inside the window. Whatever does not fit simply does not exist for the model at that moment.
The window is measured in tokens. A token (the unit the model counts in) is a chunk of text, roughly a piece of a word: a short word is one token, a long or rare one is split into several. You do not need to count tokens by hand. You just need the idea: the bigger the window, the more material the model sees at once and the less it loses the thread.
How much can Claude hold at once
The Claude 5 family, for example Claude Fable 5.1, works with a context window of up to one million tokens. That is stated in Anthropic's release notes and explained on the context windows page in their docs. A million tokens is a lot: a full mid-sized book, a year of project correspondence, a whole folder of contracts and annexes at once.
For you as a business owner this means one simple thing. You do not have to cut a document into pieces or retell in your own words what is already in the file. Load the whole thing and work with the complete text.
What this changes in real tasks
Contracts. Drop in the entire contract with annexes and ask where the hidden penalties, auto-renewal, and bad deadlines are. Claude holds the whole text, not just the first pages, so it finds the clause on page forty that you would have skimmed past.
Reports and analysis. Load a full year of sales data and last year's plan. The model compares the numbers inside one window, with no retelling and no lost detail.
Long projects. A long thread with Claude on a single task no longer falls apart: early decisions stay in the window, and you are not re-explaining the background every morning.
Does that mean size never matters again
No, and here is why. Even a large window is finite. If you dump absolutely everything into one chat, the important parts drown in the noise and answer quality drops. Order beats a pile.
Three simple rules:
- Keep in the chat what belongs to the current task, and remove old files you no longer need.
- Start a new chat for each big new topic so the model does not drag someone else's context along.
- If you reuse the same documents across tasks, move them into Projects, where files and instructions attach automatically. We have a separate write-up on projects.
Where to start
Take one document you usually wrestle with by hand: a contract, a brief, a long report. Load it whole and ask three specific questions about the substance. You will see the difference between summarizing a fragment and working with the full text in five minutes.
Want the mechanics on your own tasks rather than in theory: we have free materials (/guides) and a free first lesson about your digital twin (/try/b0-01-unit). You will walk out with a working skill, not just having read about one.
AGINE Academy is an independent product and is not affiliated with Anthropic. Claude belongs to Anthropic.
Questions
A token is a small chunk of text, roughly a piece of a word. A short word is one token, a long or rare one is split into several. Tokens are the unit used to measure how much a model processes at once.
A lot: a full mid-sized book, a year of project correspondence, or a folder of contracts with annexes. The exact figure depends on the language and formatting, but the order of magnitude is hundreds of pages, not dozens.
Better not to. Even a large window is finite, and when the important parts drown in noise, answer quality drops. Keep in the chat only what belongs to the current task, and start a new chat for each big new topic.
A context window lives within a single conversation and clears when you start a new chat. To attach documents and instructions to different tasks automatically, you move them into Claude Projects.