I downloaded a version of that openassistant model for llama.cpp and it’s at least on par with GPT-3 or a little beyond. It’s to the level of being generally useful.