Principally

1, you should perform energy minimization as it will remove steric clashes/bad interactions which arises during model generation.

2, Yes, Normally we do optimization with these steps and good structure is when you have proper stereochemical parameters, no steric clashes and no residue in outlier region or should be within accepted range. You may wish to visit 

http://nihserver.mbi.ucla.edu/SAVES/Info.php

http://swissmodel.expasy.org/docs/structure_assessment

3, Clustal Omega is good and is in most practice. 

Hello,

Prior to model a protein, we also need to predict and see what secondary features are present within the structure. I'd suggest you to first predict the structure through PSIPRED server. That should give a crude image of how your protein might actually look like. 

Post modeling your protein in modeller, you should go for structure evaluation using RC plot and other tools such as verify3d, errat to ensure the quality of your modeled protein. Check the regions that have loops for refinement. Then after you should go for energy minimization that will remove any such steric hindrance or clashes that might have happened while modeling of the side chains of the backbone atoms. Like Amarjeet Kumar said it'll all eventually depend upon the homology or sequence similarity between the template structure and query sequence that you took.

You can then proceed for simulation whether or not you intend to dock the protein with another biomolecule or a ligand. This will eventually help you in getting a stable structure and also identifying structural changes, if any. 

Thanks and Happy Research :)

Thanks everybody for your replies. 

My query sequence has 98% similarity with the template, so the model looks almost the same as the template. What I'm confused about is if Modeller's models require energy minimization, since it says on the manual that it already performs minimization steps. Also, my protein was modeled based on a template bound to a ligand, so isnt energy minimization going to "collapse" the active site?

You can use ModellerJ for protein homology modelling. Its an easy to use interface for Modeller. See this video

https://www.youtube.com/watch?v=mBgqGpztcmM&t=27s