Can the same prompt produce different answers from different LLMs?

Yes, and Centralpoint logs which model produced which answer for every inference call. This matters for reproducibility, for cost-vs-quality experimentation, and for audit when an answer is questioned later. The model is a first-class field in the inference log.