How much a model can consider at once.
WHAT IT IS
The amount of text a model can take into account in one exchange: your input plus its output.
Measured in tokens, roughly corresponding to word fragments.
WHY IT MATTERS
Everything the model knows about your specific situation must fit inside it.
Beyond that limit, earlier content is dropped or the request fails.
WHAT THIS MEANS PRACTICALLY
A long conversation eventually loses its beginning A very long document may not fit Instructions given early may fade in a long exchange
HOW CONTEXT SIZES HAVE CHANGED
Early models handled a few paragraphs. Current ones handle substantial documents.
That has widened what is practical considerably.
WHAT STILL GOES WRONG
Models attend unevenly across a very long context. Material in the middle receives less weight than material at the start or end.
Placing important instructions at the beginning and the end helps.
PRACTICAL ADVICE
Put the most important instruction first Restate critical constraints at the end for long inputs Split very large documents and process in sections Start a fresh conversation when the current one has drifted
