Jina ColBERT v2 is an advanced late-interaction retrieval model built on the ColBERT architecture, supporting 89 languages with superior retrieval performance, user-controllable output dimensions, and processing capabilities for text lengths up to 8192 tokens. It represents a revolutionary advancement in information retrieval, approximating the joint query-document attention of cross-encoders through late-interaction scoring, while maintaining inference efficiency close to traditional dense retrieval models.