Model composition enables users to combine multiple custom key-value models with a custom classification model to create a single ensemble model with a single model identifier. This is useful when the user has trained several custom models for different document types and wants to group them to analyze any of those types with a single API call.

The model alias feature allows users to define a friendly name to refer to a custom model in code rather than using the model identifier. If a new version of the custom model is created later, the new version can be assigned the alias originally given to the earlier version and the code that invokes the model does not have to be changed.