Repository navigation
Support GPT-5.6 family prompt caching without changing older models #36315
Description
Description
Summary
OpenAI added new prompt caching controls for GPT-5.6 and later models.
We should support those new options without changing the existing behavior for GPT-5.5 and older models.
Why this matters
GPT-5.6 uses prompt_cache_key for more reliable cache matching, and supports request-wide cache settings through prompt_cache_options.
Older models should keep using the current behavior. The OpenAI docs say older models reject the new prompt_cache_options and prompt_cache_breakpoint fields, so we should only send the new fields when they are explicitly configured.
What should change
Add support for passing this through OpenAI provider options:
providerOptions: {
openai: {
promptCacheKey: "session-or-cache-key",
promptCacheOptions: {
mode: "implicit",
ttl: "30m"
}
}
}
Activity
github-actions commentedon Jul 10, 2026
This issue doesn't fully meet our contributing guidelines.
What needs to be fixed:
[FEATURE]:title prefixPlease edit this issue to address the above within 2 hours, or it will be automatically closed.
This issue might be a duplicate of existing issues. Please check:
prompt_cache_key— same pattern of adding provider-specific prompt cache key/options supportprompt_cache_key— same concept of per-provider prompt cache key supportIf you believe this was flagged incorrectly, please let a maintainer know.
Sun-ZhenXing commentedon Aug 19, 2026
When using GPT 5.6 Luna, I occasionally encounter the following message:
prompt_cache_retention is not supported on this model. This led to an interruption, I think it might be relevant.