Yes, I think you're best off to use HB. Importances can be distorted by aggregation if you use the other models. For example, assume a two-level color attribute. Half your respondents have a utility of +100 for the color orange and -100 for pink. Unfortunately, the other half of your respondents have a utility of -100 for orange and and +100 for pink. In an aggregate logit model these utilities will exactly cancel out and you'll have a utility of 0 for orange and 0 for pink. In your importance calculation you will also have a 0 importance for color.
But clearly color IS important to your respondents - they just disagree on which colors are better than others, half preferring pink and half preferring orange. In HB this color preference heterogeneity will come out. Importances are calculated at the respondent level and then aggregated so color can have a significant importance even if average utilities are 0 for orange and 0 for pink.