context window

Reading Time: 5 minutesInput, output, cached: three meters run every AI call. How caching, model routing and context tiers cut a developer’s token bill without hurting quality.