AI summaryⓘ
The authors created jina-reranker-v3.5, a smarter and faster tool that helps rank lists of documents for better search results, especially when the data is a mix of different types like legal, medical, and financial texts. They improved how the model pays attention to parts of the text by combining sliding-window and global layers, which makes it more efficient without losing important comparisons. They also made the model learn better by using a step-by-step teaching method where a bigger model guides a smaller, faster one. This new model performs as well as much larger models but with fewer resources and is particularly better at handling semi-structured data. The authors also shared the model publicly but only for non-commercial use.
listwise rerankingcross-document comparisonsliding-window attentionglobal attentionself-distillationmulti-domain trainingnDCG@10semi-structured datainference latencylanguage models
Authors
Christina Nasika, Feng Wang, Antonis Krasakis, Han Xiao
Abstract
Listwise rerankers are the discriminative core of agentic retrieval pipelines, yet production deployment demands efficiency, domain robustness, and fluency on semi-structured data at the same time. We present jina-reranker-v3.5, a 0.6B-parameter listwise reranker that meets these demands together without sacrificing the cross-document comparison that makes its predecessor jina-reranker-v3 effective. jina-reranker-v3.5 keeps the last-but-not-late (LBNL) interaction of jina-reranker-v3 and reworks it along three axes. It replaces uniform global attention with a hybrid schedule of three sliding-window layers followed by two global layers, pinning the terminal layer to global as LBNL readout requires. It trains on a curated multi-domain mixture that spans legal, medical, financial, multilingual, and structured retrieval. It transfers quality through a three-stage self-distillation recipe in which a full-attention teacher sets an upper bound that a sparse-attention student then recovers under a staged adaptation protocol. jina-reranker-v3.5 reaches 63.20 nDCG@10 on BEIR, matching a 4B model at roughly 7x fewer parameters, and improves over jina-reranker-v3 on MIRACL and RTEB as well. Its largest gains come on semi-structured retrieval, where it lifts nDCG@10 by 9.6 points over jina-reranker-v3 and leads all rerankers of comparable size. The hybrid schedule further cuts listwise inference latency by up to 1.56x. We release the model weights on Hugging Face under a non-commercial license.