Running multiple LLMs sounds smart because different models have unique...
https://quebeck-wiki.win/index.php/The_Real_Economics_of_Tokenization:_Why_Output_Costs_More_Than_Input
Running multiple LLMs sounds smart because different models have unique strengths and failure modes. You get to compare output and catch errors. But watch out. Synthesis often hides vital dissent or spreads your private data across more vendors