No standard conjoint method, whether it involves OLS regression, logit, latent class MNL, or HB MNL allows direct comparison of the part-worth (dummy-coded for categorical attribute labels) utility of one level from one attribute to the utility of one level of a different level. This is due to the experimental design, where one and only one level from each attribute is brought together to form a product concept in the conjoint questionnaire.
You CAN compare the utilities of levels WITHIN the same attribute, allowing you to judge relative preference within an attribute.
You can also conclude a consumer's preference regarding joint combinations of attribute levels. For example, you could sum the two part-worth utilities for Red and Convertible (levels from two different attributes) and compare it to the sum of Blue and Sedan to see which sum of utilities is higher (and therefore which product concept the respondent would be more likely to choose).