Science/Anatomy models for educational purposes worth producing for selling?

Medical visualizations is a very lucrative category to create for but you need to avoid creating things that are already there.

My bet would rest on detailed models at the cell level, there are less of these because they are harder to make.

Note harder to make things have benefit you can set higher price on them and they just sell.

If you can make good quality models, insane good renders (very important), nice presentation and detailed description, support multiple formats and make models that are in demand and “not there jet”, then you would be surprised how much you can earn.

As iterate says (so tempted to say I'm going to reiterate lol), if you can make decent quality and detailed models for medical usage then you can set a good price for them and $600 isn't such a distant goal to have.

If you are indeed going for biological models I would also recommend you to learn 3D Coat, its a voxel based sculpting program ideal for creating the complex organic forms usually associated with bio models. Your going to generate very high poly meshes with it but you can retopologize them in 3D Coat itself, 3ds max, Blender, etc.

3ds max is a good software for all kinds of stuff but not that good for fast organic free-form modeling, in that regard Blender actually has the edge over 3ds max.

Blender provides a very powerful toolset, if you can work well with it then basically you do not need 3ds max. I consider apps like 3ds max and Blender sort of a workbench, its good for getting all the 3D/2D data in and assemble stuff, animate it and render things out, but they do not always provide best tools to generate the 3D/2D assets themselves.

For long time main power of 3ds max was its parametric nature (modifier stack), the polygonal modeling tools and animation capability's, but 3ds max has not kept pace with competition and is starting to get somewhat outdated now in many way's.

Purchased Houdini indie license some time ago (for just 230 dollars on Steam) and it just just keeps amazing me all the time, so much power for that little money, mind blowing software. Its more oriented towards producing things procedurally (and non destructively) and that's just so much more convenient in many way’s (especially for medical visualizations/effects) so also consider taking a look at Houdini.

Regarding the polycount,
there is no case to be made on how many that needs to be, in general just do not add polys that do not contribute to overall shape. For game or real-time application it all depends on the use case and the target compute budget for given application, a developer could for example opt for low grade phone hardware and thus need extremely low poly models but some developer targeting next gen PC and GPU hardware can opt for more detailed models, etc.

Regarding file conversions,
we usually provide OBJ and FBX format (every major 3D app can export/import them) “if it only involves models”. If it involves specific render setups and scenes including animation controllers, etc., then one needs to import own models into target software (one wants to support) and redo some parts that could not be transferred, then save the scene to native format of that app.

Also consider Alembic format for animation and recently USD (universal scene description)
(https://graphics.pixar.com/usd/docs/index.html), 3D industry is now finally ready to adopt a common scene description for exchanging 3D scene data, Blender, Unreal engine and Houdini already support it and soon probably every major 3D app.

Brilliant answer iterate! Nothing more to say.