Anyone here tried both Qwen and GLM families on the same setup and found a clear winner for one task vs the other?