AI glossary
What is a context window?
The context window is the maximum number of tokens a model can take into account at once: your message, the conversation, any documents and its own answer.
It works like the model's working memory in a conversation. Anything inside the window it can use; anything left outside does not exist for the model.
It is measured in tokens, not words or messages. Current models have windows of hundreds of thousands of tokens, and some reach a million. But fitting in does not guarantee the model pays equal attention to everything: in very long texts, details in the middle can get lost.
When a conversation goes past the limit, the app has to trim or summarize the oldest part. That is why a long chat sometimes "forgets" something you told it at the start.
An example
If you paste a very long contract and it does not fit in the window, the model will answer without having read part of it.