How does this compare with the existing open source AndroMDA project on SourceForge?

The GMT project emphasises model transformations as "first class model citizens". This distinguishes the GMT objectives from most other current MDA tools. Current MDA tools (including the current capabilities of GMT) only provide direct PIM to textual source code transformations, usually via a template language. The two main issues with the state-of-the-art are: (a) there is no [OMG] standardised template language, and (b) mapping straight from a PIM to textual source code at a much lower level of abstraction leads to less than elegant template code. The OMG QVT (Query, Views and Transformations) RFP addresses the issue of standardisation of model transformations, but the work in this area has only just begun. GMT will allow the definition of PIM-to-PSM transformations in a practically useable syntax, which will lead to much simpler PSM-to-text transformation template code.