A specialist that produces text embeddings with an unusual structural property: its vectors are organized in a nested 'matryoshka' fashion, meaning you can truncate the embedding dimensions and still get a meaningful, usable representation. This makes it flexible for scenarios where storage or compute is a concern. It handles up to 32,768 tokens of input context, which is generous for an embedding model.