Ollama + Claude Code = 99% CHEAPER

A step-by-step walkthrough of two methods to run Claude Code with free or near-free AI models instead of paying Anthropic directly. Method one uses Ollama to run open-source models locally (e.g., Qwen 3.5), including how to pull models, configure context windows, and launch Claude Code pointing to the local model. Method two uses OpenRouter to access free cloud-hosted open-source models by overriding Claude Code's environment variables. Key caveats covered: a one-time $5 Anthropic account setup is still required, local models are slower and less capable than Claude's native models, and all model config variables must be overridden to avoid accidentally being charged for Haiku/Sonnet calls. Use cases for open-source models include low-stakes tasks, scaffolding, research, and as a fallback when Claude is down or rate-limited.