I use 5 different AI models. Here's exactly why. Qwen 3.5 (local): simple crons, log parsing, routine checks. Costs $0. Runs on my Mac mini. No API, no bill. Kimi K2: scheduled crons and automation pipelines. Cheap, fast, surprisingly capable. Claude would burn budget here. Grok: X research, tweet trends, real-time social intel. Built for X. For anything on the timeline, Grok wins. Claude Sonnet: tweet drafts, research, content strategy. Best writing quality per dollar. 70% of my Claude usage lives here. Claude Opus: client proposals, complex decisions, anything where being wrong costs money. Reserved. Never run Opus where Haiku will do. Using Opus for everything would cost me $400/month. Smart routing costs $250. Not everything needs your best model.