firstname.lastname@example.org Your model comparison is of the two models against each other (yellow and blue lines) and with the Dual Lift you could see how they perform against actual values as well. From the MASE scores both of the models are about 14%-15% better than the Baseline or naiive model.
For accuracy and consistent prediction, the AVG Blender is sligthly better then xGboost, but not significantly, when looking a the lift chart. xGboost lift chart is not monotonically increasing. However, the slightly better accuracy has a cost. AVG Blender is almost twice as slow as xGboost.