i've been running qwen 3.6 35B A3B with llama.cpp on a 3090ti. i have found it better then sonnet in many ways. Speed and iterations was key. here is the gist of my current configuration: https://gist.github.com/spencerkittleson/5e44b6895a17ca45161... I use this with tailscale so all my devices have full access to it. That machine get toasty....