The OpenAI API now supports 12-hour prompt caching, allowing developers to store shared instructions, tools, and context for extended periods. Requests made hours apart can still be billed at the lower cached input rate, targeting continuously running agents that reuse context throughout the day. Additionally, cache data for logged-in users of Codex and Work will be retained permanently. When users return to or continue an old conversation after several hours, they will hit the existing cache without incurring additional costs.