What's this about?
OpenAI’s Thibault Sottiaux says the default grows from 272K to 500K, about 84%, to reduce compaction during long tasks. The model maximum remains approximately 1M; a larger working context is not a guarantee of perfect recall or lower total cost.
What exactly is context window?
How much text a model can consider during a request; compaction summarizes older material to make room.
What's the community saying / why does it matter?
The announcement had 7,319 likes; no independent evaluation was collected. This changes a product default, rather than announcing a new model.