Offline Agentic Coding part 3: Apple Silicon costs more than OpenRouter.
Summary
This article analyzes the cost and performance implications of running local LLM inference on Apple Silicon versus OpenRouter hardware. It covers power usage, electricity costs, price per million tokens, and throughput, concluding that hardware costs dominate while local inference can be cheaper under certain scenarios but remains slower than cloud options in some cases.