Mac mini M5 Pro users discuss practical local AI setups
A Lobsters thread asks which local models, quantizations, and runtimes are genuinely usable on a Mac mini M5 Pro with 48GB of unified memory. Commenters say Qwen3.8-27B is their preferred local coding model despite being slow and limited for long contexts or parallel sessions, while another user runs Qwen3.6-27B at 6-bit through MLX-LM for routine tasks and codebase scanning.
The thread offers first-person guidance on where 48GB unified memory is useful for local coding, assistants, ComfyUI, and image generation, while showing that practical usefulness can depend more on latency and workflow than on whether a model technically runs.
What can I realistically do with local AI on a Mac mini M5 Pro (48GB)?
authored by

napolux2026-11-08 11:07:51 | 2 comments
#Mac mini M5 Pro#Qwen3.8-27B#Qwen3.6-27B#gemma4-e4b#MLX-LM#Claude#DeepSeek 4.1 Flash#ComfyUI