12-Hour Prompt Caching

For developers on the OpenAI API: repeated prompt content can now stay cached for 12 hours instead of 30 minutes, so requests spread across a workday still get the cheaper rate.

Details

Caching means OpenAI remembers the opening part of a prompt it has already seen and charges much less for it on later requests. Until now that memory lasted 30 minutes by default
You switch to the longer window with one setting (prompt_cache_options.ttl set to "12h"), or leave the 30-minute default in place
One request can mix both: 12 hours for shared instructions that rarely change, 30 minutes for newer conversation content
Saving into the 12-hour cache costs 2.5x the normal input rate, versus 1.25x for 30 minutes. Reading from either costs the same, 0.1x the normal rate on most models
Works on GPT-5.6 and later models in the Responses API only. 12 hours is a minimum lifetime, not a guaranteed cache hit