OpenAI
OpenAI's 2h Window Took Five Minutes and Used Its Own Cache
OpenAI's two-hour completion window gives 30% off for waiting. In my runs the wait was five minutes, a reused continuation_id cut it to seconds, and the window kept a prompt cache of its own: a 7,426-token prefix that was warm on Standard was written again in full by the first 2h request.
21 Nov 20267 min read