Why mostly unusable 125b?

I assume you are talking about qwen3.8-flash-next. Support for it on some places, like llama.cpp, is still wip (depending on configuration) but it looks like a very capable model in it's category.