No, that is actually a new feature of the OpenAI embedding model. You have the ability to selectively discard or drop certain dimensions, and the model will still function appropriately.
Under the hood is Matryoshka Representation Learning aniketrege.github.io/blog/2024/mrl