From Nature: https://www.nature.com/articles/s41586-024-08252-9#Fig3
“As shown in the scorecard of Fig. 3, the forecasts of GenCast are significantly more skilful (P < 0.05) than that of ENS on 97.2% of our 1,320 variable, lead time and vertical level combinations (and 99.6% of targets at lead times greater than 36 h)“
Seems like a significant improvement on its face, would leave the more technical analysis of the graphic to those more knowledgeable