LLM Local Inference Sheet

Which local LLM and runtime to run on a MacBook, measured on an Apple M3 Pro (18 GB).

• Results dashboard — speed, quality, capacity and compatibility
• Proofreading on a Mac — Qwen3-8B vs Qwen3-4B for grammar correction and tone rewrites
• Technical Kitchen — how the numbers are measured and how far to trust them
• Repository — code, data and findings