Would I be surprised there’s bench maxing happening? Yes. But some users also use Q4 quantized and complain how dumb local models are.