Is there a way to see in a create call response whether or not a fallback model was used? I've been seeing some weird results with GPT 4.1 and thinking it might be falling back to some other model.