Claude Code Cost Control Starts with Model Routing

The strongest takeaway is that cost control comes from routing intelligence where it matters, not blindly downgrading every task. The claimed 90% savings will vary by workload, but the workflow is practical.

• –opusplan reserves Opus for architecture and planning while using Sonnet for implementation.
• –Subagents can inherit the main model or be explicitly routed to cheaper models.
• –Prompt-cache hits are dramatically cheaper than fresh input, making cache reuse crucial for long coding sessions.
• –/model switching is useful, but model changes can affect cache reuse and erase some expected savings.
• –Developers should measure actual token and cache usage instead of assuming the cheapest model always delivers the lowest total cost.