Comment by mark_l_watson

I have been running the slightly larger 31B model for local coding:

ollama launch claude --model qwen3.6:35b-a3b-nvfp4

This has been optimized for Apple Silicon and runs well on a 32G ram system. Local models are getting better!