Recently, the Standard Performance Evaluation Corporation (SPEC) unveiled the latest advancements in its AI benchmark, SPEC ML. This benchmark introduces three pivotal metrics to evaluate the performance, scalability, and, notably, model-arithmetic efficiency of software and hardware systems across diverse AI workloads. Notably, the inclusion of model-arithmetic efficiency in the SPEC ML benchmark evaluation marks a significant milestone, addressing a critical gap in the assessment of large model computation efficiency. As AI technology continues to permeate various sectors, the establishment of a robust benchmark testing system has gained unprecedented importance. Arthur Kang, the chair of the SPEC ML Benchmark Committee, emphasized that a standardized benchmark evaluation approach not only facilitates streamlined model comparisons but also fosters innovations that prioritize efficiency, accuracy, and sustainability.