I just use both. One is doing review work the other is doing planning and coding. I pick which one does which arbitrarily. I have zero proof that the end result is better either way around. But I tend to have the reviewer on "high" and the coder on "medium" for whichever model my instinct thinks is best for the task. I think it is really hard to "prove" which one is better at this point as so much is dependent on AGENT.md/CLAUD.md, prior documentation/plans the code etc. If someone tells me one is better than the other I take it with a grain of salt. When dumb things happen I simply blame both of them - they are team afterall.