{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiblyad2f5gmbshpln3srk5jkrf5ymvhndu6t5gzsnkizhtzsyia64",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mkl27g7lzpp2"
  },
  "path": "/t/switching-to-gpt5-4-nano-results-in-0-cache-hit-rate/1379973#post_1",
  "publishedAt": "2026-04-28T16:32:40.000Z",
  "site": "https://community.openai.com",
  "textContent": "While running a test of 20 back-to-back identical prompts, with ~1450 token system prompt prefix, I am consistently getting > 90% cache hit rate when using gpt-5.4-mini. Simply switching the model to gpt-5.4-nano and running the same test always returns a 0% hit rate even after multiple runs.\n\nI am not aware of any difference between the prompt caching capabilities of these models so I do not understand why switching to 5.4-nano would result in a 0% cache hit rate.",
  "title": "Switching to gpt5.4-nano results in 0% cache hit rate"
}