Key Info

OpenAI has extended prompt caching in its API to a 12-hour window, and for logged-in users across Codex and Work the cache is kept indefinitely with no penalty for returning to an old conversation.

Highlights

  • API users can now select a 12-hour prompt cache duration.
  • Logged-in Codex and Work users keep cached context permanently.
  • Resuming a conversation after hours incurs no penalty and is now guaranteed to hit the cache.
  • Longer-lived caching suits persistent agents that reuse shared instructions, tools, and context throughout the day.
  • Cached input is billed at the lower cached-input rate even when requests are hours apart.