Inside the eval gate: how we decide a cheaper model can take over a task

I build Operant, so I have a bias. This post first went out on our blog. I post it here because I...