Prompt Caching Not Working for GPT-5.4-Nano

The Prompt Caching docs make it sound like prompt caching should work out of the box for GPT-5.4-Nano, yet subsequent identical request to the Responses API show no change in the number of cached tokens. I’m seeing multiple unresolved discussions about the same thing online. Since I can’t add links in a post, Google “prompt caching not working gpt-5.4-nano” and there are at least 3 discussions about this. Are we all missing something here? Thanks