Don't trust concept art perspective to be true all the time. It may lie to you.
Anyway, I think what's important is breaking things down. Dumb advise but it's just true.
On this asset I don't think it curves on this plane
Your's droops and curves, I think it's way more planar. I also don't think it sticks out from the side of the gun as far as you have it. Sorry don't have much advise, imo it's just a case of trying to understand some planes and relationships with other parts and then developing from there.
Exspanding upon zachagreg's well thought out examples that really is a key aspect to think about or in other words pre-planning prior to modeling anything, will mostly mitigate potential problems from occurring, any how a few additional points to keep in mind:
1. Where possible I'd advise at this stage to mainly model off references, rather than using a free style 'eyeballing' method which requires a ton of practice and experience to capture a near enough copy of the subject so for this exercise I drew a 'blueprint' from your posted pics upthread and analyzing the main shape we'll start with a plane:
2. Now couple of things to note, if while sketching your model extra edge loops are necessary to outline a detailed profile, then by all means add them, because sufficient geometry is a main consideration to support a well designed topology, well at least from my experience. Also transformation action constrain too axis will confine either vertical or horizontal segments alignment, therefore minimising mesh surface 'peaks and valleys' a typical cause of shading artifacts.
4. With the mirror modifier applied and we'll then validate/check that our edges are properly aligned along their applicable axis then hopefully the object should as expected shade correctly once sub-divided (lvl 3)
5. In my opinion seems fine plus importantly via GI preview mode, no noticable shading problems.
If you're unsure, focus on the silhouette. Switch the shading to a flat color (with no texture) as needed while modeling to determine if the details/polygons you're putting in affect the overall silhouette (when the model is at the expected distance from the player/viewer), and if they don't feel free to discard them.
Like Tiles said, triangles is not the main concern when it comes to performance. For an open world, the landscape usually takes up a huge chunk of performance and memory due to it covering so much of the screen while usually being fairly heavy on the shader side (lots of texture blending) .
A great starting point so to think about how the model will be used. Is it a gun for COD that you're supposed to inspect and enjoy all the juice details? Or is it a gun that's always gonna be held by a character in third person? So what's important for the game and how will it be viewed.
Tying in to that is silhouette. From an aesthetic point of view, silhouette should drive where you put your triangles.
With all that said, depending on engine and shader a lot of times I will add support edges to create a better normal bake. I will put in extra splits in my model to be able to UV better. And in production some of that just flies out the door because it's a low importance prop and time is more important. Oh and then there's modeling so that you can LOD better, meaning modelling so that you can UV better.
In the end all of this will come with experience. Stick to application and most triangle bang for the buck and you're golden.
When somebody asks me such a question then my advice is usually to start with some much smaller projects first. There is no such chart. Every game is unique. That's one of the first things that you learn.
The "trick" that you search for is called prototype. There is no better method than to simply try it out. There are too many factors that influences the game performance and needs. A good starting point is similar projects. Have a look there what poly count they use, what shaders, what map types etc. . But then you need to try it out if it fits to your project. What works for them might fail for you.
The general modeling rule applies also to game asset needs: as much geometry as needed, as few as possible. And when you come close then you need more detail. But the pure poly count is never the problem. The performance bottlenecks usually arrives at other things like Shaders, particle systems, game logic, texture map types and resolution, how many characters at once at the screen. The environment. The game type. The engine limits. The target platform limits. And so on.
So make a prototype, then have a look at the performance. Then you can decide if you can go up in some details, or if you need to go down in some details. And take also into account that more detail usually requires more man power. There's a reason why so many flat shaded comic style games are around nowadays.
I decided to take part in this challenge. Here is my progress so far. I haven't made game ready assets before so trying my best to keep it low poly not entirely sure if what I am doing is correct or should I just carry on modelling high poly then work backwards and clean up/retopolgise?