> One option to get the costs down would be to use a local LLM It's not that cheaper if you want similar level quality and inference speed. For commercial usag

> One option to get the costs down would be to use a local LLM It's not that cheaper if you want similar level quality and inference speed. For commercial usage, OP will require training his own model since no good commercial alternative exists.