Seems the prompt caching is broken (at least for GPT 5.5 and 5.4) It used to work a few days ago. Does anyone else have the same problems?
Hi and welcome back!
I just tried and couldn’t reproduce this. In all my tests caching was observed.
Are you still seeing this issue?
| Model | Requests | Errors | Cached tokens sequence | Result |
|---|---|---|---|---|
gpt-5.4-2026-03-05 |
4 | 0 | [0, 10496, 10496, 10496] |
Caching observed |
gpt-5.4-mini-2026-03-17 |
4 | 0 | [0, 10496, 10496, 10496] |
Caching observed |
gpt-5.4-nano-2026-03-17 |
4 | 0 | [0, 9984, 9984, 9984] |
Caching observed |
gpt-5.5-2026-04-23 |
4 | 0 | [0, 10496, 10496, 10496] |
Caching observed |
Hi,
I got an email notification from OpenAI status page (in early hours for me). So I don’t know for sure, but this might be related to the recent OpenAI status incident with elevated 431 errors across API components. Since Chat Completions and Responses were among the affected components, it could have impacted request handling around the same time. The status page says it’s resolved now, so if caching is still missing after recovery, it may be a separate issue.
Thanks for raising this @georaven7. I can see why this was confusing, especially if prompt caching had been working normally for you just a few days earlier.
Also appreciate @VeitB for checking and sharing their test results, and thanks to @LarisaHaster for pointing out the status page incident around the same timeframe.
This issue has since been resolved. If you're still seeing missing or inconsistent prompt caching behavior, please create a support ticket by emailing support@openai.com with as much detail as possible (model used, timestamps, request examples, screenshots/logs if available).
Once you receive a case ID, feel free to share it here so the team can help monitor and troubleshoot further if needed.
-Mark G.
Many thanks! 10014807 - this is my case