@ni_nad I very briefly tried running a distilled version of Qwen3 Coder on an M4 MacBook with Ollama but found the experience pretty underwhelming. I hear good things about Kimi K2 in the open weights model department but haven't looked too much into it.

From what I've read, the only viable machines for local LLMs are overspecced MacBooks and Mac Minis with their unified memory, or the NVIDIA DGX Spark which is purpose built for inference.