The context window is how much text a model can hold in view while it works. Through 2026 a million tokens became routine on serious models, at no extra charge – something that would have been unthinkable a year earlier.
How much is that
Roughly, a million tokens is several thick books. Or a mid-sized code repository. Or a year of email correspondence.
What it enables in practice
The first change is that you no longer have to cut documents into pieces. Working with a long text used to mean splitting it, processing each part separately, and stitching the results together, a process that loses the connections between parts. Now the model sees the whole thing.
The second is working with code. An agent holding an entire project in context understands how the parts lean on each other. That is the difference between a change that works and a change that breaks something else.
The third is long-running interaction. An agent working a task for hours does not lose what it did at the start.
Why bigger is not always better
First, it costs. Input tokens are billed, and if you send the whole document on every call, the bill grows linearly. Prompt caching softens this considerably when part of the context is identical between calls – the repeated portion is charged at a much lower rate.
Second, it is slower. More context means longer waits for a response.
Third and most important: attention is not evenly distributed. Models still use the beginning and end of the context most reliably. A fact buried in the middle of half a million tokens can be missed. This has improved, but it has not gone away.
A practical rule
Do not dump everything in just because it fits. Send what is relevant, and put the most important material at the start or the end. A model given ten pages of the right material gives a better answer than one given five hundred pages containing those ten.
A large context is the answer when you genuinely need the whole thing – a full contract, a full project. It is not a substitute for thinking about what the model actually needs.