OpenAI now lets developers choose a 12-hour prompt cache on the Responses API. OpenAIDevs announced the option at 01:56 Beijing time on October 21, and the OpenAI API changelog lists the change under October 20.
For GPT-5.6 and later, the prompt caching guide says to set prompt_cache_options.ttl to "12h" for a 12-hour minimum lifetime. Cache writes at that lifetime cost 2.5× the ordinary input-token price. The default "30m" lifetime writes at 1.25×. Cache reads use the same reduced cached-input rate for both lifetimes, so a hit still pays the lower cached-input price either way.
The changelog says 12-hour prompt caching keeps shared instructions, tools and context cached longer so requests that are hours apart can still benefit from lower cached-input pricing. OpenAIDevs also pointed to persistent agents that reuse context throughout the day as an example of when the longer lifetime helps.
The documentation ties the "12h" and "30m" settings, and the matching write rates, to GPT-5.6 and later on the Responses API. Absolute dollar prices for specific models are not restated here; October 2026 pricing uses the multipliers above against each model's ordinary input rate.