OpenAI Adds 12-Hour Prompt Caching Option to API

Developers can now extend prompt cache retention to 12 hours instead of the default 30 minutes to maintain persistent context across spaced requests.

Developers can select 12h retention, keep the 30-minute default, or mix both durations in a single API request.
Requests can apply 12-hour caching to shared instructions and 30-minute caching to recent conversation context.
Cached prompts are billed at the lower cached-input rate even when calls are hours apart.

Developers running persistent agents with large shared contexts can cut input token costs across multi-hour query gaps.

Sources