What changed #
Usually ignored rather than rejected, so the call succeeds and does something subtly different.
Affects prompt_cache_options on CreateModelResponseProperties reachable from 2 operations. Grep for it: nothing else will tell you.
Who is affected #
Code that sends OpenAI's CreateModelResponseProperties objects, used by 2 OpenAI operations.
Before and after #
| Where | Before | After |
|---|---|---|
| CreateModelResponseProperties.prompt_cache_options | present | absent |
| BetaCreateModelResponseProperties.prompt_cache_options | present | absent |
Code that breaks #
await client.post("/…", {
// prompt_cache_options is now ignored, not rejected
prompt_cache_options: value,
});
Based on OpenAI’s published API specification. How severity is decided.
source & further reading
driftwatch.et — original article
Vercel: type can now return "project-expiration-removed"
[Additive] Vercel: justification is now optional
OpenAI: value can now return "max_output_tokens"