Cascade routing tries the cheap model first and escalates to a more expensive model on low confidence. The break-even (vs always-expensive) is at what cheap-model success rate?
Answer: >50%
The math: cascade total cost = cheap_share × cheap + (1 - cheap_share) × (cheap + expensive). For this to be less than always-expensive: cheap_share > 50%. Below 50%, the doubled-call cost on failures exceeds the savings on successes.