K-Culture GlossaryㅋWhere everyone starts
Context Window
The amount of material an AI can "hold in mind" and read at once. When it's large, you can feed in an entire book.
In plain words
A context window is the maximum amount an AI can "remember and read" in a single conversation. Think of it like a desk: a big desk lets you spread out several books and compare them side by side, while a small one forces you to clear away the previous page just to look at the next.
It's measured in tokens. An announcement of "1 million tokens of context" means you can feed in the equivalent of several books at once and ask questions about them. That's what makes it possible to drop in a lengthy contract, a thick report, or an entire novel and ask, "Summarize this," or "Find the contradictions."
The reason a chatbot seems to "forget" the start of a long conversation is also about the size of this window. Once content falls outside the window, it's pushed out and effectively no longer exists for the AI. That's why every new model release touts how many times bigger its context has become.
How it shows up in the news
"They expanded the context window to 2 million tokens" — meaning the amount of material that can be read at once has grown by that much, often mentioned alongside uses like analyzing long documents or understanding an entire codebase.
Try it yourself
- Pick a long document you have (PDF or Word) and attach it in a ChatGPT or Claude chat window.
- Type: "Summarize this document in 5 lines, and list separately any parts that contain numbers."
- Getting an answer based on the whole document is the power of the context window at work. If you try asking about something from early in a very long document, you can also feel its limits.
Context Window vs. Memory Feature: What's the Difference?
How much you can read at once, Whether it persists after the conversation ends
Both get described as "the AI remembers," but one is only valid within the current conversation while the other carries over to the next one. Forgetting the beginning of a long conversation is a context window issue; the AI knowing your name today after you mentioned it yesterday is a memory feature.
| Aspect | Context Window | Memory |
|---|---|---|
| How long it lasts | Only within this chat session. It disappears when you close the window | Persists into the next conversation, even the next month |
| What it is | A limit on how much can be input at once | A feature that stores information separately and retrieves it later |
| When it overflows | The earlier parts get pushed out and forgotten | If too much has accumulated, some memories get deleted or selectively used |
| What's debated | It becomes a numbers race — marketing around things like 1 million tokens | Privacy. What is okay for it to remember |
| Analogy | Documents you can spread out on a desk at once | Files kept in a drawer |
Rule of thumbIf it disappears when you close the chat window, it's the context window; if the AI still knows it in a new conversation, it's the memory feature.
See also
Stories using this term
No story has used this term yet. New ones attach here automatically.