OpenAI API adds 12-hour prompt caching; cache writes cost 2.5x input price
OpenAI's October 20, 2026 API changelog adds an optional 12-hour prompt cache for v1/responses. 12-hour cache writes cost 2.5x the input token price. Later requests can still get lower cached-input pricing, even when they arrive hours apart.
SessionWatcher editorial · Published · Updated · Source announcement: 2026-10-20
What OpenAI changed on October 20
OpenAI's API changelog entry dated October 20, 2026, listed under v1/responses, says developers can now choose 12-hour prompt caching in the API. The entry presents it as an option that developers turn on.
OpenAI says the longer cache keeps shared instructions, tools and context cached longer. Requests that arrive hours apart can then still get the lower cached-input price instead of paying full input pricing again.
Sources: Openai release notes: 2026-10-20
What it means for your API bill
The extra cost comes when content is written to the cache. OpenAI says 12-hour cache writes cost 2.5x the input token price. Later requests that reuse the cached content are billed at cached-input pricing, which the entry describes as lower.
The changelog entry does not give the cached-input rate, per-model prices or the write price for shorter caches. From this text alone you cannot work out how many cache hits it takes to make up for the 2.5x write cost.
Sources: Openai release notes: 2026-10-20
What the entry does not cover
This is per-token API pricing. The changelog entry does not mention Codex, ChatGPT subscriptions or plan usage limits. It does not say whether those products use 12-hour caching or whether it affects their allowances.
Sources: Openai release notes: 2026-10-20
AI assisted reporting, checked against the linked official sources. Source pages checked 2026-10-20. Editorial process and corrections.