As a modeler I'd like to just pipe in and reiterate some of the statements already made here as well as give a bit more clarification on some issues raised.
The release of our models, and a step-by-step creation process, is highly doubtful to ever be released public for various reasons. Mainly, the team as a whole have figured everything out on our own as we'll expect anyone motivated enough to do as well. We did all the work and we don't want people just freeloading off our efforts. Secondly, and the most important reason, these models that we've created are part of our personal portfolio as well as the coders have created proprietary code they'll use for their own portfolios as well.
I, personally, don't spend 2-weeks to a month modeling an asset just to give it away as public domain. A model I create goes into my demo reel and job applications and giving it public would allow someone to copy, modify or reuse something I spent many man-hours working on. Artists sadly do this all the time, steal artwork, intentionally or unintentionally and use it as their own. There are many models available online to download and dissect, but are not catered to our workflow.
Second, any model I make for this project, based on the NDA I signed when I joined, cannot be distributed publicly because I waved all rights to the model and its uses to MWLL and Wandering Samurai Studio. Once I create that model and upload it to the SVN it's not my model any more and belongs to the game which waves my distribution rights so even if I wanted to give it away, I'm not allowed to do so.
Now, for some general information, nothing to specific but a solid idea of how our models are created and processed from the perspective of a modeler. Generally, you can use any package to model but the final product must be realized in 3DSMax to assign shaders and smoothing groups, as well as exporting of materials, bones and animations. Crytek only released 3DS toolsets so I can do the modeling in Maya, but usually export as an OBJ or FBX to import to 3DS then continue finalization from there.
General polycount is variable, preferably less than 10k polys per model, the less the better in terms of overall performance, but if you're going for less polys, you must emphasize your lack of geometric detail with such things as normal maps, bump maps, specular maps, ambient occlusion maps, diffuse maps and so forth. The normal maps are the biggest thing to net the most detail from any model while retaining the highest possible performance.
As an example, the Puma model is somewhere between 1000-2000polys while the mesh the normal map was rendered from (detail texture) is about 13million polys. That's how the low-poly retains the illusion of depth rendered by the engine. Variably speaking, most models range from 2000-7000polys with a max of 10k and have 1-3 LevelOfDisplay models (LOD's). Greebling your model (adding misc details) is what the high-poly is for, while the low poly is generally for shape and form. Every polygon counts as you only model what needs to be modeled for it to function and react physically with the environment.
Some models share the same basic properties, such as a 3-bone foot (3 separate parts) to create articulation. Legs themselves are two-parts, calf and thigh with a center pelvic base to connect legs and torso. Torso is one single part, with modular extensions added afterthefact aside from arm parts such as canon geometry for certain chassis (like the complex arm geometry of the Madcat and Thor). The cockpit is modeled directly into the interior of the mesh, inherent in its actual design, so we have to take considerations of canon design versus functional design (as in, can you actually see out of this cockpit, can we change it enough to actually have a decent visual range while retaining the integrity of the canon design).
Above all else, I at least strive to make sure there is no clipping and each limb has a min-max free-range of motion so during its maximum animation state the limbs do not clip through any particular piece of intersecting geometry. General rules of poly modeling come into play, for example, since many of these are hard-body objects there doesn't need to be a full-focus on making sure every polygon is 4-sided (general rule of thumb for an animation pipeline), triangles are accepted in only quantities integral to the shape of the geometry you need to create. The engine triangulates everything anyway but quads are win since it makes UV Mapping and Texturing easier. Triangles, as a general rule-of-thumb for modelers, are BAD for characters due to the negative effect they have while skinning (vertex weights) as they create pinching and tearing. This is less of an issue for mechs and vehicles.
These are some general criteria and thought-processes while modeling an asset for the game pipeline. Since I can't get into specifics, I hope this answered some of the questions raised in this thread and gives those willing to model a better idea of what we do when working on asset creation from a modelers perspective.
