Tag
GLM-5, a 744B parameter model, underperforms on the MMLU-Pro benchmark compared to a much smaller 9B model, raising questions about efficiency and scaling.