Chinese models like Qwen? Can those be run on local hardware (like a 4090) with good results? I thought the appeal of services like Claude was that you needed bazillions of memory for good results. I’m very out of the loop.