Interesting. My anecdotal experience using multiple LLMs at once was mostly a disaster, despite the code being simple it seemed like when I used the same code and fed it to a new LLM, the new LLM would break basic things repeatedly.