The context window
The context window is how much text the model can see at once, fill it and quality degrades, so start a fresh chat when it runs low.
The context window is everything the model can see right now
Every time the model generates a reply, it works from a fixed-size window of text, the context window. Anything inside that window the model can use. Anything outside it, the model cannot see at all.
Context is measured in tokens. A token is roughly 3–4 characters of ordinary English text, so a token is usually shorter than a word. "context window" is two tokens. A typical page of text is around 500–700 tokens. Current models have context windows ranging from around 100,000 to 200,000 tokens, enough to hold a very long conversation, but not unlimited.
Fill the context window
Add items below and watch the window fill. When it overflows, the oldest messages are dropped, the model no longer sees them.
What fills the window
The context window is not just your latest message. It holds everything the model receives for this turn: any instructions you or the app has set (system prompt), every message in this conversation, both yours and the AI's replies, and any files or documents you have attached.
That "context usage" percentage or bar some tools show is telling you what fraction of the total window is occupied. At 20%, you have plenty of room. At 90%, the window is nearly full.
When the window fills, one of two things happens depending on the model and app: either the oldest messages are silently dropped so new ones can fit, or the model starts to lose coherence as it tries to compress too much into too little space. Either way, very long chats tend to get worse, the model forgets what you said at the start, loses the thread, or starts giving generic answers.
Question 1 of 4
What counts toward your context usage?