I would test this, might be cheaper per task even costing more per token, probably faster too

I believe the older models are being gradually phased out, newer ones have no availability issues