Mac mini M5 Pro owner asks what local AI actually works on 48GB

The post is a practical ask from someone who has just bought a Mac mini with an M5 Pro (base CPU config) and 48GB of unified memory. The owner frames the problem clearly: it is hard to separate what actually works day to day from benchmark hype. That framing drives the questions — which models and quantizations fit comfortably in 48GB and are actually usable, not just "it runs." The owner also asks which runtime and tools to use, and what others gave up on and returned to cloud APIs for. The request for real tokens/sec numbers and "waste of time" stories suggests the owner values hands-on experience over spec sheets. The scope covers coding autocomplete and agentic workflows, image generation, and other genuinely useful tasks like transcription and RAG over notes.