1. A quick comparison

The fastest way to see two climbers on the same problem is to name both when you start a run:

hillclimb run circle-packing --budget 30m --climber greedy --climber openevolve

That starts one search per climber, in one run, each tagged with the climber it uses. Follow them like any other searches, and compare them when they are done:

hillclimb watch                            # the searches side by side
hillclimb chart                            # both climbs on one chart
hillclimb experiment report <run-id>       # the two compared in a table

--climber takes a preset (greedy, openevolve, gepa) or a Python file of your own.

Keep it fair

Searches that run at the same time share the folder's memory, so what one learns can reach the other halfway through. For a fair comparison turn memory off:

hillclimb run circle-packing --budget 30m --climber greedy --climber climbers/mine.py --no-learning

Both searches also share the machine's coding agent slots. If the machine is busy, one may wait for a slot while its clock runs. Give each search room, or compare them with an experiment, which can start searches one at a time.

When it is not enough

A quick comparison is one search per climber. It is good for a first look, for checking that your climber runs at all, or for spotting a very large difference. It cannot tell you whether a small difference is real, because it does not know how much a single search varies on its own. For that, write an experiment: 2. Writing an experiment.