Django ORM uses the active record implementation – you’ll see this implementation in most ORMs. Basically what it means is that each row in the database is directly mapped to an object in the code and vice versa. ORM frameworks such as Django won’t require predefining the schema to use the properties in the code. You just use them, as the framework can ‘understand’ the structure by looking at the database schema. Also, you can just save the record to the database, as it’s mapped to a specific row in the table.
SQLAlchemy uses the Data Mapper implementation – When using this kind of implementation, there is a separation between the database structure and the objects structure (they are not 1:1 as in the Active Record implementation). In most cases, you’ll have to use another persistence layer to keep interact with the database (for example, to save the object). So you can’t just call the save() method as you can when using the Active Record implementation (which is a con) but on the other hand, you code doesn’t have to know the entire relational structure in the database to function, as there is no direct relationship between the code and the database.
So which of them wins this battle? None. It depends on what you’re trying to accomplish. It’s my believe that if your application is a mostly a CRUD (Create, Read, Update, Delete) application which no hard and complex rules to apply on the relationships between the different data entities, you should use the Active Record implementation (Django). It will allow you to easily and quickly set up an MVP for your product, without any hassle. If you have a lot of “business rules” and restrictions in your applications, you might be better with the Data Mapper model, as it won’t tie you up and force you to think strictly as Active Record does.