Introducing the Ettin Reranker Family

Six new Sentence Transformers CrossEncoder models, known as the Ettin Reranker family, have been released in various sizes. These models are designed for "retrieve-then-rerank" pipelines, where a fast embedding model first retrieves candidate documents and the reranker then accurately re-orders them based on relevance. Built on ModernBERT encoders, the models were created using a distillation recipe, and the full training process and data are shared. The rerankers work by taking a query-document pair and outputting a single relevance score, which is more accurate but computationally more expensive than using embeddings alone.