We have moved from the Chat Completion API to the Response API for gpt-5 and gpt-5-mini.
A few of our use cases requires HTML output.
With the Chat Completion API, HTML language output works well for both Latin and non-Latin languages such as Japanese, Korean and the Hindi languages. With the Response API, HTML language output is completely garbled for all non-Latin as well as some Latin based languages.
Correct BCP 47 language tags are present in the HTML as well as UTF-8.
We are at a loss and at our wits end on this.
Is anyone experiencing the same issue? Any suggestions? Or is this a serious OpenAI bug that needs to be fixed?
TIA
Edit: The instruction and input prompts are both written in English.