Smart Routing — Cheap vs Reasoning
Not every task needs Opus. Routing the boring 80% to a fast model cuts cost and latency without losing quality.
Every turn with Claude Code costs tokens. Opus 4.7 is brilliant and expensive. Sonnet 4.6 is fast and cheap. Routing simple tasks (rename a variable, write a small test, format a file) to Sonnet and reserving Opus for hard reasoning (architecture, multi-file refactors, debugging gnarly bugs) cuts your daily spend by 40-70% with no quality loss.
Tools like RuFlow extend Claude Code into a 60+ agent system where smart routing is built in. The routing decision uses signals like: estimated diff size, whether the task involves cross-file reasoning, prior failure on this kind of task, and explicit overrides in CLAUDE.md.
The simplest version you can run today: keep Opus as your default, but add a CLAUDE.md note that says 'For one-file edits under 50 lines, prefer Sonnet.' The agent reads that and routes appropriately. You'll be surprised how often Sonnet is enough.