Hi Ignacio,

I agree that there're many cases that you'd use an open source solution instead of OpenAI's API, especially when you need to fine-tune a LLM model on private data, and Mixtral 8x7B could be a good candidate for the job; but for general use, calling the API is the easiest way and probably most cost effective way, that's why OpenAI may take a lead in this field for a long time. We know hosting the API incurs enormous cost, OpenAI has deep pockets and can bear the cost without worrying about losing money for now, but I'm not sure that'd last forever, and if the competitors could do the same.