NEWVectors or files. Pick a path.Start →
    Models/Reranking/jinaai/jina-reranker-m0
    HFRerankingcc-by-nc-4.0

    jina-reranker-m0

    by jinaai

    Multimodal reranker handling text, images, and mixed documents across 29 languages

    123Kdl/month
    120likes
    2.4Bparams
    Identifiers
    Model ID
    jinaai/jina-reranker-m0
    Feature URI
    mixpeek://reranker@v1/jina_reranker_m0_v1

    Overview

    Jina Reranker M0 is the first production-grade multimodal reranker from Jina AI, handling text-to-text, text-to-image, image-to-text, and text-to-mixed-document reranking in a single model. Built on Qwen2-VL-2B-Instruct, it supports 29+ languages and up to 4K image resolution with dynamic patching.

    On Mixpeek, Jina Reranker M0 serves as a universal second-stage reranker for any retrieval pipeline: whether the candidates are text documents, scanned pages, product images, or mixed content. Its 91.02 nDCG@5 on ViDoRe v1 makes it state-of-the-art for visual document reranking.

    Architecture

    Cross-encoder based on Qwen2-VL-2B-Instruct. 2.4B parameters. Dynamic image patching up to 4K resolution. Outputs relevance scores for text, image, and mixed-modality inputs. 29+ language support.

    Mixpeek SDK Integration

    // Reranking is a retriever STAGE in Mixpeek, not an ingest-time extractor.
    // The rerank stage runs a cross-encoder inference service; the shipped default
    // is BAAI/bge-reranker-v2-m3. Pointing it at jina-reranker-m0 means registering that
    // model as a custom reranker plugin and naming it in feature_uri, which is an
    // Enterprise path. Stage contract read from GET /v1/discovery/stages.
    const retriever = await mx.retrievers.create({
      namespace_id: "my-namespace",
      retriever_name: "search-then-rerank",
      stages: [
        {
          stage_name: "candidates",
          stage_id: "feature_search",
          parameters: { limit: 100 },
        },
        {
          stage_name: "rerank_results",
          stage_id: "rerank",
          parameters: {
            inference_name: "BAAI__bge_reranker_v2_m3",
            query: "{{INPUT.query}}",
            document_field: "content",
            top_k: 10,
          },
        },
      ],
    });

    Capabilities

    • SOTA visual document reranking (91.02 nDCG@5 on ViDoRe v1)
    • Text, image, and mixed-modality reranking in one model
    • 29+ language support including CJK, Arabic, and European languages
    • Dynamic 4K image resolution for high-detail document pages
    • Code retrieval support (63.55 nDCG@10 on CoIR)

    Use Cases on Mixpeek

    Visual document search: rerank scanned PDF pages by layout+content relevance
    Multilingual retrieval: rerank across 29 languages with one model
    Product search: rerank by visual and textual product attributes
    Code retrieval: rerank code snippets and documentation together

    Benchmarks

    DatasetMetricScoreSource
    ViDoRe v1 (visual doc)nDCG@591.02Jina AI, 2026: Model Card
    BEIR (text-to-text)nDCG@1058.95Jina AI, 2026: Model Card
    MIRACL (18 langs)nDCG@1066.75Jina AI, 2026: Model Card

    Performance

    Input SizeText + Image (up to 4K resolution)
    GPU Latency~12ms / pair (A100)
    GPU Throughput~80 pairs/sec (A100)
    GPU Memory~6 GB

    Specification

    FrameworkHF
    Organizationjinaai
    FeatureReranking
    OutputRelevance score per candidate
    Modalities
    RetrieverCross-Modal Reranker
    Parameters2.4B
    Licensecc-by-nc-4.0
    Downloads/mo123K
    Likes120

    Research Paper

    Jina Reranker M0

    arxiv.org

    Build a pipeline with jina-reranker-m0

    Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

    Run it on your own data, free