I haven't tested programming tasks with a local LLM vs. say, Claude 3.5. But it is nice to be able to run 14-32B LLMs locally and get an instant response. I have a single 3090.