There has been some discussion about this lately. There are still many unknown variables on which the modelling depends. E.g. will it be possible to attach objects (sprites/models) to single bones by script? If not, animated objects like the shovel, the bow etc. will have to have animations for many actions which favours to have them as models. But otherwise, things like barrels which are simply shown as being held in the hands of the clonk could be sprites.
And, if we had (automatic) LOD (which is part of OGRE - but at the end we don't use it), then we just create a models that look good for 64x64 and ingame, a model with less polygons is shown. If someone will implement LOD is not known either.

So, independent of if they will be sprites or if there will be LOD + models, I wanted to clarify one thing:
The idea of this post is to remind the modelers that their models (for small, collectible objects) will never be shown in a higher resolution than 64x64 pixels (as title graphic) and 20x20 ingame anyway. So anything else than color maps would be overkill.
I wanted to silence the voices which said the models look too blocky when looking at a 300x300px preview of the model as well as those who said that X polygons is far too much. The guideline continues to be "so many polygons that the geometry looks good when displayed in the ingame-size" and for these objects it is 64x64.

For now, we will be using sprites since without LOD, using models for small objects is not really optimal.