Comment by quietsegfault

It’s more than just data locality. OpenRouter is faster, no? I have an M4 pro, and anything but the smallest dumbest models are unusably slow for interactive use. I personally haven’t yet found a good use case for offline/non-interactive LLM work locally.