I run a Gemma 4 32b abliteration (int8) and it's remarkably good. It's been a real step up from Qwen in my experience.