Briefing

Local means the weights stay here

A local model runs on your CPU or GPU. You pay in memory and speed, not per token. A small model will miss tasks a large hosted model gets right. The first move is a smaller job, not a bigger prompt.