Claude Code is one of the best agent coding tools out there (whether you’re vibe coding or you’re a seasoned dev), but the pricing is the worst part—especially once you’re looking at $100–$200/month plans. In this video I show you how I route Claude Code CLI’s inference to cheaper models, so you keep all the Claude Code “bells and whistles” while paying a fraction of the cost. The main swap I’m doing is replacing Sonnet/Opus with GLM 4.7 from Z.AI, which my community has been recommending a lot, and mapping Haiku to GLM 4.5 Air. I walk through the exact Claude Code setup: editing your .claude/settings.json to add the env routing block, setting the API base URL to Z.AI’s Anthropic-compatible wrapper, and mapping Claude model names to GLM models behind the scenes. I also show the right way to handle API key security using settings.local.json plus a .gitignore entry, and I explain the difference between project-level vs global user-level settings. Finally, I demo running Claude Code with GLM models (you’ll literally see GLM 4.7 show up), how to skip the annoying onboarding bug by flipping hasCompletedOnboarding in your claude.json, and where to check usage inside Z.AI’s API management screen.