Glossary
Context Window
Context window — The maximum amount of text, counted in tokens, that a model can consider at once when producing its next reply.
In more detail
The context window is a hard budget covering everything the model can see: system instructions, the character definition, the visible conversation, and the reply currently being written. When the total exceeds the limit, something has to be dropped.
Products handle that differently. Most drop the oldest messages silently. Some summarise what they drop, preserving the substance while losing the exact wording. Many pin the character definition so personality survives even when specifics do not.
Larger windows are not automatically better. They cost more per message, which is why generous windows and generous free tiers rarely coexist, and models attend unevenly across a long window — material at the beginning and end typically carries more weight than material in the middle.
Treat a published window size as a ceiling on what is possible rather than a promise about what will be remembered.