Think of it as a desk, not a memory
A context window is the total amount of text an AI model can look at at one time. Not store forever, look at right now, in this exact reply. It includes your original prompt, anything you have pasted in, every message back and forth in the conversation, and the answer the model is about to write.
Picture a desk with a fixed amount of space. Every new page you add takes up room. The desk does not grow. So once it is full, adding a new page means an old one falls off the back. The model is not deciding what to forget. It simply cannot see anything that does not fit on the desk anymore.
Why it feels like forgetting
Early in a conversation, everything fits easily, so the model tracks details perfectly. You mention your name, your project, your preferences, and it uses all of it naturally. That is because the whole conversation still fits inside the window.
Keep chatting long enough, though, and the earliest messages get pushed out to make room for the newest ones. Ask about something you mentioned at the start, and the model has no way to answer correctly, because that part of the conversation is no longer in front of it. It is not a bug. It is the desk running out of room.
This is also why long documents get cut off
The same limit explains a different frustration: pasting in a long document and getting an answer that only seems to cover part of it. If the document, plus your question, plus the model's answer, add up to more than the window allows, something has to give.
Sometimes the tool will simply refuse the whole thing. Other times it quietly works with only part of what you sent, which is worse, because you may not realize anything was left out.
How big is the window, really?
Every model has a different size, and the sizes keep growing as models improve. That is exactly why we are not going to hand you a specific number here. Whatever we said today would likely be out of date within a year, and different models have very different limits at any given time.
What matters is the habit, not the number. Assume every model has some limit, and that a conversation or document long enough will eventually reach it. Check the specific tool you are using if the exact size matters for what you are doing.
How to work with the limit instead of fighting it
A few habits make this a non-issue for almost everything you do.
- → If a conversation has drifted far from where it started, begin a new one rather than dragging the old context along.
- → For anything long-running, summarize the key points so far in one message. This resets what matters without losing it.
- → Put the most important facts near the start of your prompt, not buried at the end of a long document.
- → If a document is huge, consider splitting it into sections and asking about each one separately.