This result really surprised us too, but it seems like we were not alone in observing this: https://medium.com/@gabrielgrinberg/openai-api-fine-tuned-gp...

That being said, both accounts that we experimented on were personal accounts ( with a very limited budget) so maybe there are some special behaviors (e.g. some rate limit that only exists for base models)

Yeah we were definitely surprised! This was a small scale experiment; we uploaded our notebook for latency tests here if anyone is interested taking a look at the code / trying it their own finetuned models: https://github.com/eqian99/finetuning-experiment

Under the same conditions where enterprise versions of the API have significantly less latency and better reliability than personal. OpenAI can change anything about the underlying infrastructure.