Cut Your Claude Code Costs: How Token Caching and Smart Habits Save You Money

Maximizing the value of your Claude Code sessions

Claude Code's token usage and costs are driven by input/output token pricing, prompt caching, and how much context accumulates in a session. Output tokens cost ~5x input, and thinking tokens are controlled by effort level. Prompt caching automatically stores conversation prefixes, making re-reads cheap (0.1x), but changes to model, effort, fast mode, or timeouts invalidate the cache, causing expensive re-prefills. To minimize costs, keep context short and relevant: @-mention files, use quiet command flags, /clear between tasks, /compact when done, and /rewind instead of /compact for recent mistakes. Also, start fresh sessions for loops and avoid switching models mid-conversation.