I'm currently running experimental version of llama.cpp in order to get the new Qwen3.8-Flash-Next to run. It's running incredibly well. We'll see what it produces.