So difficult to read.

In my opinion, Harness has nothing to do with energy (ie cost). They all use the same conversational requests so the efficiency is mostly in the prompt (instruction and data) as well as the chosen model.

Choose wisely.

For efficiency, providers should cache the initial prefix (ie system prompt and tools) of harnesses, not sure if they do all that.