The short answer
The context window is how much text a model can hold in view while it answers: your conversation, anything you attached, and its own replies. It is not memory. Nothing is stored between sessions unless a separate feature saves it. When a conversation grows past the window, the earliest part drops out of view, which is the moment it starts to forget.
It is a desk, not a filing cabinet
Picture a desk with a fixed amount of space. Everything the model can consider right now has to be laid out on it: your question, the conversation so far, any document you pasted, and the answer it is writing.
When the desk is full, something has to come off. Usually the oldest material goes first. That is why a long conversation starts losing what you said at the beginning: the name you specified, the constraint you set, the tone you asked for.
And when the session ends, the desk is cleared completely. Start a new conversation and it has no idea who you are or what you were doing, unless the product has a separate memory feature that quietly puts some of it back.
DOT / THE TRANSLATOR
This is the single most common misunderstanding I meet. People think they are talking to something that remembers them. They are talking to something that is handed a transcript each time and reads it very quickly.
Memory is a product feature, not a property of the model
Several tools now claim to remember you between conversations. That is real, but it works differently from what people assume: the product stores notes about you separately and pastes the relevant ones back onto the desk at the start of each session.
This matters for two reasons. It is a stored record of you that exists somewhere, which you can usually inspect and delete. And it consumes desk space, so the remembering costs you room that could have held your actual question.
What to actually do
- Start a fresh conversation when you change subject. Old context helps nothing and crowds the desk.
- Re-state the important constraint if a long chat starts drifting; it has probably fallen off the desk.
- Look for the memory settings in any tool you use often, and read what it has decided to remember.
- Paste the part of the document that matters rather than the whole thing, when you can.
Why a bigger window is not automatically better
Model makers advertise ever larger windows, and it sounds like an unambiguous improvement. It is not.
Attention thins out. Material in the middle of a very long input gets noticeably less weight than material at either end. Burying the key sentence in the middle of two hundred pages is a poor way to be heard.
It costs more, too. Everything on the desk is billed as input, every single turn, so a large window used carelessly is a large bill arriving quietly.
And a big window invites bad habits: pasting everything rather than choosing. A well-chosen five pages usually beats a lazily dumped five hundred.
BEEP / THE MYTH-BUSTER
Yes, I can read your entire inbox. No, that does not mean I will notice the one sentence that mattered. I would rather you handed me the three emails than the filing cabinet.
See it for yourself
Reading about it only gets you so far. This one is written to show you the thing described above, rather than describe it again.
I am about to paste something long. Before you read it, tell me what you will do if it does not all fit, and whether you would mention it or just quietly drop the beginning.
Then paste it into whichever assistant you use — ChatGPT, Claude, Gemini or Copilot. Not sure which? Meet them all.
If you remember nothing else
- The context window is what the model can see right now, not what it remembers.
- Long conversations lose their oldest details, which is when it starts forgetting.
- Between-session memory is a separate stored record you can usually inspect and delete.
- Material in the middle of a very long input gets the least attention.
- Everything in the window is billed every turn, so bigger is not free.
Where this came from
- Current recorded context sizes, per model, with their dateschecked 26 September 2026
Anything here that turns out to be wrong is worth telling us about. Figures that move — prices, sizes, energy use — are kept in the registry with their dates, rather than written into the prose where they would quietly go stale.
Questions about “Context windows”Asked by readers, answered by us. Open to read them.
Open to load them.
Ask about this page and we will answer it here, where the next person can read it too.
Nobody else can answer you, and you cannot answer anybody else. This is not a forum — we would rather say that than run a room we cannot watch at three in the morning.
Your question goes up with the answer and without your name. Nothing identifies you, ever.
If we cannot answer it we will tell you why, and that one stays between us.