Even the exact same model isn't enough. There are several sources of nondeterminism in LLMs. These would all need to be squashed or seeded - which as far as I know isn't a feature that openai / anthropic / etc provide.