GLM-5.2 Becomes the Top Open-Weights Model: Active vs Total Parameters

GLM-5.2, an open-weights model by Zhipu AI, has topped the Artificial Analysis Intelligence Index. This is significant because it highlights the difference between active and total parameters, which affects a model's memory footprint and compute costs. To understand a model's costs, consider both total (744B) and active (40B) parameters. This distinction is crucial for sparse models like Mixture-of-Experts, where not all weights are used for every token. As a result, engineers should pay attention to both numbers when evaluating model releases.