If you look at axial capacity by itself it's at about 40% of capacity. That's the code check the program is reporting for uniaxial load. Look at a detail report and you will see exactly where this falls on the interaction diagram. Think of this code check as the distance between the load point and the top of the interaction surface.
But, for bi-axial load, the program is using the PCA load contour method which looks that the axial load and sees how much moment can be added to it to bring the column to capacity. Think of this as the distance between the load point and the SIDE of the interaction surface.
In that sense, both code checks are accurate. However, that being said, it would make more sense if the program (when it uses PCA Load contour) would report the worst case of the Bi-Axial PCA load contour result and the pure axial load case!! I'll put in a change request to have the program do that. That would make more sense to the user.