OpenAI API adds 12-hour prompt caching with higher write pricing
Longer caching covers supported Responses API models; 12 hours is not a deletion deadline or a guarantee of savings.
OpenAI announced available 12-hour prompt caching on October 20, 2026, for GPT-5.6 and later models in the Responses API. Total write rates are 2.5× standard input pricing, versus 1.25× for the default 30 minutes, not surcharges on ordinary input. Twelve hours is a minimum cache lifetime, not a deletion deadline. Savings depend on reuse; faster responses are not guaranteed. The exact effective time was not provided. This is not a ChatGPT web or mobile feature. No additional Traditional Chinese support was announced. Not tested here.
Usage conditions
- Hong Kong
- Hong Kong is absent from the official OpenAI API supported-region list, so local use of this API feature is unsupported. This is not a ChatGPT web or mobile feature. No regional bypass is provided.
- Macau
- Macau is absent from the official OpenAI API supported-region list, so local use of this API feature is unsupported. This is not a ChatGPT web or mobile feature. No regional bypass is provided.
- Taiwan
- Taiwan is API-supported, subject to an eligible API account, access to supported models and usage-based charges. Account-specific model access and limits require checking. This is not a ChatGPT web or mobile update; Traditional Chinese performance was not tested.
Official source: OpenAI API changelog — 12-hour prompt caching