Can you double check that my call to the new gpt-4-0125-preview is indeed using that model? OpenAI has claimed on its blog that it has fixed UTF-8 characteres for non-English languages, however my clients are still getting this:
(I'm just another Helicone user not staff FYI) but maybe just try it through the OpenAI Playground - the model reporting in my experience is just exactly what the request was, so I suspect it's correct in Helicone (based on my experience) - but easy to verify using the OpenAI Playground directly and see if you get the same result