宝玉 (@dotey) · 10/21/2026, 13:17:21
OpenAI Enhances Codex and API Caching with Permanent Retention for Logged-in Users
OpenAI has announced significant updates to its prompt caching mechanisms for Codex, Work, and the API. For logged-in users of Codex and Work, cache retention is now permanent, ensuring cache hits when resuming old conversations without penalty fees. Additionally, API prompt cache retention can be extended up to 12 hours (default remains 30 minutes), with first-time writes charged at 2.5x the standard rate and subsequent reads costing only 10%.
SOURCE COVERAGEFull Original Report
这些天以来目前为止最有价值更新：Codex 和 Work 的登录用户，缓存会永久保留，回到几小时前的旧对话或恢复会话，都保证命中缓存，没有“惩罚”（我理解是不会按照缓存失效收费）。
怀疑是保存到磁盘了，那也很牛👍
除此之外，OpenAI 开发者官方账号宣布，API 的提示缓存现在可以保留 12 小时，默认仍是 30 分钟。首次写入收原价 2.5 倍，之后读取缓存只收 1 成。
↳ 引用 @thsottiaux (Tibo): Day 16/
You can now cache for 12h in the API and for logged in users across Codex & Work we keep cache forever and you don’t have any penalty for coming back to an old conversation or resuming after a few hours and are now guaranteed to hit the cache.
