https://store-images.s-microsoft.com/image/apps.10812.8dc4e9ba-6331-41ad-9757-ebb2e79f8499.4cf864a1-cb4b-4b7b-bada-5166612348c1.afcfce50-248f-4775-9648-d294c5ba6563

Jina Reranker v1 Turbo - en

Jina AI

Jina Reranker v1 Turbo - en

Jina AI

A state-of-the-art fast neural text reranking model supporting 8192 sequence length.

  • Jina Reranker v1 Turbo model is a neural text reranking model, designed to enhance the relevance of search results.
  • This model is the best balance between accuracy and performance, offering fast and more memory-efficient reranking process.
  • For our most accurate (and larger) reranker model, please see Jina Reranker v1 Base - en.
  • Jina Reranker v1 Turbo complements text embedding models and refines search results by prioritizing documents relevant to a query.
  • This state-of-the-art reranker model enables a variety of applications that rely on precise search results, improved information retrieval, and better data organization.
  • Use-cases: Vector search, retrieval augmented generation.
  • See our embedding models (Jina Embeddings v2) on Azure for state-of-the-art 8k embedding models for vector search.

Hightlights:
  • Trained for speed and accuracy: While performing only slighly worse than Jina Reranker v1 Base - en on some of the benchmarks, this model can process three times as many documents in the same time.
  • Extended context length: This reranker model is capable of handling queries up to 512 tokens and documents as large as 8192 tokens.
  • High performance across the board: This reranking model ranks at the top compared to its competitors, in terms of 'Mean Reciprocal Rank' (MRR), according to BIER, MTEB, LoCo and an independent benchmark by LlamaIndex. A higher MRR represents a higher chance that the most relevant document to a query is returned with the highest relevance score by a reranking model.