The guys working on OData4j, http://code.google.com/p/odata4j/ , have done some amazing work with their Java version of an OData Consumer API.  I would recommend that the new Objective-C API use a Fluent API, http://en.wikipedia.org/wiki/Fluent_interface, which would make it simpler to quickly get up and running without the need and overhead of generating entities.  Like I said I really enjoy using the OData4j API with its great design and its simplicity.

+1

And I'd also like to integrate directly to Core.Data.

Also, it'd be nice to make the library easily substitutable to enable unit testing of apps. It should ship 2 implementations of the API. One is the real deal. The other operates only on an in-memory cache. So it simulates what the real system would do, starting from a server with an empty database and without any network overhead.

Mostly, I want to eliminate the code gen. That is a very C# approach, and feels absolutely bizarre to the platform (besides being annoying overhead).

I completely agree and the Core Data integration would be very nice.