PRODUCTION SCENARIO
Ridgeway Compliance runs a GPT-5.6 review job every two hours; each run sends 30 requests that share a 6,000-token instruction block and finishes in about ten minutes. Usage shows cached_tokens on every request after the first in a run, and never on the first.
Why does the first request of every run miss the cache?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: A cached prefix stays eligible for only 30 minutes after its last write or reuse
On GPT-5.6 and later a cached prefix stays eligible for reuse for 30 minutes after its most recent write or reuse, and OpenAI only may keep it longer, so a two-hour gap lets the entry lapse and the first request writes it again. The block is well above the 1,024-token minimum, and no cache key is needed on these models.
Official Source: https://developers.openai.com/api/docs/guides/prompt-caching ↗