Card #2858 · OpenAI Platform & APIs
Prompt cache hit rate dropped to zero for volatile context windows
The user is experiencing a sudden and complete loss of prompt cache hits despite no significant changes to their architecture. Previously, the system handled volatile context windows effectively, but now hit rates have dropped to zero, leading to increased costs.
Pain score
0.45/ 1.00 · weighted product of four components
Reach0.32distinct authors (log-normalized)
Recency0.85freshness decay, half-life months
Engagement0.14upvotes + comments, normalized
Monetization1.00willingness-to-pay cues in evidence
§ Evidence — 3 verified quotes
my ability to hit cache at all has dropped to zero
I’m unable to get access to the same cache mechanisms as before
My absolute best efforts to reverse engineer the black box mechanics of their prompt caching mechanisms have managed to alleviate some of the costs, but you can see pretty clearly that today’s post-fix fingerprint is still not significantly improved.
Comments are a Pro feature — join the conversation, share what you've shipped, hear what others are building.
Upgrade to Pro