Celestia's new CMOD format is like 3DS in that it's just another way of defining a 3D shape. It's much more efficient because Chris designed it to match the data structures actually used in 3D graphics cards. Celestia's frame rate seems to be between 1.5 and 2 times faster when displaying a CMOD model than when displaying an equivalent 3DS model of the same object.
There really aren't any GUI modelers yet that can generate CMOD models. The squares in the graticule actually were done by hand when I realized that coding around the "fencepost" problems would take longer, but I wrote a trivial g77 Fortran program to generate the circles and fiducial marks. (I suspect Fridger would have used a perl script

) I understand that Chris has been using some conversion utilities, but I don't think they're freeware.
Sorry, I don't know of any way to turn CMOD models off and on at runtime. They're just like 3DS models in that regard. Facets can have normals and be invisible when you look at them from the wrong side, but the graticule consists only of lines -- I don't know of any way to associate normals with them. But I'm certainly no 3D expert.
Christophe has created a DOxygen Web page from the text in the new Mesh and Model source code files. It's at
http://celestia.teyssier.org/source-documentation/classAsciiModelLoader.html
Bear in mind that this is still in beta test. Models created now might not work at all with future pre-releases or in 1.3.2 final.
I hope Chris doesn't mind, but...
I've added a section to my "cryptic Celestia Notes" page that includes some of the things he's written on the developers' mailing list as well as some of the comments from the source code which describe some of the features of the new model format. See
http://www.lns.cornell.edu/~seb/celestia/celestia_notes.html#14.0