Context window
How much the model can hold in view at once — your prompt, the docs, the whole chat — before the start falls off.
Think of it like
Short-term memory, or a desk that only fits so many papers before the older ones slide off the edge.
Example
Paste a 400-page PDF into a small-context model and it will quietly lose the early chapters.
How it actually works
Windows have grown from a few thousand tokens to millions, but “it fits” is not “it uses well.” Models often attend better to the beginning and end than the muddled middle — the classic “lost in the middle” effect.
For product teams
The model's attention budget — overload it and details drop, so scope what you feed it.
For engineers
Max token span the model attends over; content beyond it is truncated or must be chunked/retrieved.
Related
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome