Client was running 500+ conversations/day through their AI agent. Burning $450/month on GPT-4. The fix I built in n8n: Added a smart routing system: 1. DeepSeek R1 classifies query complexity (costs almost nothing) 2. IF complexity_score > 7 → GPT-4 3. ELSE → GPT-4o-mini What happened: - 60% of queries went to GPT-4o-mini (way cheaper) - Only complex stuff hit GPT-4 - Same quality, zero user complaints - New cost: $120/month ROI: Took 2 hours to build, saves $330/month = $3,960/year The key insight: Most queries don't need GPT-4's full power. Route smart, save money. Anyone else optimizing API costs? What strategies are working for you?