According to CoinMeta, on the 16th day of its 28-day product sprint, OpenAI announced that the prompt cache for Codex and ChatGPT Work users would be permanently retained, and the optional cache duration for API would also be extended from the default 30 minutes to 12 hours. This means that even if users continue working after a long pause, they can still ensure cache hits, avoiding the loss of the benefits of cache reuse. In addition, the cost of the first write to the 12-hour cache is 2.5 times that of a regular input, while for subsequent cache hits, the input cost for most new models is reduced by 90%.