My guess as to what is happening is that the the gradient vectors do not have unit amplitude. If you look at this post, you’ll see that by default MRtrix3 will scale the b-values according to the amplitude of the gradient vectors (this is to support multi-shell schemes on certain scanners). You should be able to confirm using the mrinfo 14_1.mih -dwgrad -raw_dwgrad. If you’re confident that your b-values were 3000, use the -bvalue_scaling 0 option to disable this in those commands that support it.