OpenAI API adds 12-hour prompt caching for long-running agents: cache writes cost 2.5x input, reads keep the discounted rate

Why it matters

【Key points】

OpenAI added 12-hour prompt caching to its API on Oct 20. Shared instructions, tool definitions and reference material can stay cached for at least 12 hours and still bill at the lower cached-input rate when requests are hours apart.

1. How: for GPT-5.6 and later in the Responses API, set prompt_cache_options.ttl to "12h"; the default stays 30 minutes. One request can mix both, e.g. 12 hours for shared instructions and 30 minutes for newer conversation content.

2. Price: 12-hour cache writes cost 2.5x the normal input rate, versus 1.25x for 30-minute writes. Reads cost the same either way: 0.1x input on most models, 0.05x on GPT-6.1 Sol.

3. OpenAI says 12 hours is a minimum lifetime, not a deletion deadline, and does not change data-retention policy.

4. OpenAI's example is always-on agents like dots that reuse the same context all day, where savings from repeated reuse can outweigh the higher write cost.

【Verifiable】

Prompt caching guide: https://developers.openai.com/api/docs/guides/prompt-caching#cache-lifetime

API changelog: https://developers.openai.com/api/docs/changelog

OpenAI Developers post: https://x.com/OpenAIDevs/status/2112603916076781778

【Our read】

1. This is a billing option built for persistent agents, whose calls are often minutes to hours apart, so 30-minute caches keep expiring.

2. Writes cost twice as much, so it only pays off with enough reuse in a day; sparse or fast-changing workloads are better off on the 30-minute default.