OpenAI adds 12-hour prompt caching in the Responses API
A 12-hour cache keeps shared instructions and tools priced as cached input across longer gaps. Writes cost 2.5 times the normal input price. Reads stay at the usual cached rate.
Official source (opens in a new tab)
OpenAI's API changelog for 20 October 2026 says the Responses API can keep a prompt cache for 12 hours. The longer cache holds shared instructions, tools, and context so requests hours apart still pay the cached-input price. A 12-hour cache write costs 2.5 times the normal input token price.
The prompt caching guide (opens in a new tab) says this applies to GPT-5.6 and later. The default 30-minute write costs 1.25 times the normal input price. Those figures are the whole write price, not an extra charge on top of ordinary input. Reads cost 0.1 times ordinary input on most of these models, and 0.05 times on GPT-6.1 Sol, either lifetime. OpenAI may keep an entry longer than the minimum you ask for. Choosing 12 hours does not set a deletion time. Moving tokens that already have valid 30-minute coverage up to 12 hours bills a new 12-hour write, with no credit for the earlier write.
Who it's for: anyone calling the Responses API with a long prefix that is reused through the day, not only inside one short session.
Fits: Best for Vibe Coding. The Codex rank is the cloud agent and the edit loop. A cache lifetime on the API does not change that verdict.