I dont get it, so for slow inference its:

hermes > pi > opencode ??

or is hermes better in general? habent tried it hermes/pi yet