I mean we only tolerated those bills for a few months before we had to switch, but generally speaking we were handling on the order of 4-5M requests/day, mostly

I mean we only tolerated those bills for a few months before we had to switch, but generally speaking we were handling on the order of 4-5M requests/day, mostly low bandwidth. API Gateway, and wasting lots of allocated server resources for each time the function ran, was just killing us. The load was pretty constant, which was another reason to switch. No regrets either way, but we should have done the billing math at production load and not just been blinded by the free tier and the low cost per run.