Jina Reranker v2

Jina Reranker v2 is Jina AI's second-generation text reranking model (jinaai/jina-reranker-v2-base-multilingual), a full XLM-RoBERTa cross-encoder covering more than 100 languages. Unlike bi-encoder embedding models, it reads the query and each candidate passage together and emits a single relevance logit, which makes it substantially more accurate for second-stage reranking in RAG pipelines.

The model is trained on both prose and code relevance, so it can rank function signatures and code snippets against plain-English requests as well as multilingual documents. It is widely adopted as the recommended default multilingual reranker in open-source RAG tooling and is available through Jina's rerank API and Hugging Face. The release date below is era-level (2024).

License note: the weights are released under CC-BY-NC-4.0, which restricts commercial use; commercial deployments should use Jina's API or a commercial license.