jina-reranker-v3
by jinaai
Compact Qwen3-based reranker for multilingual text retrieval precision
jinaai/jina-reranker-v3mixpeek://reranker@v1/jina_reranker_v3Deploy jina-reranker-v3
Single-tenantMixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.
Overview
Jina Reranker v3 is a production text-ranking model from Jina AI built on Qwen3-0.6B. It is useful as a second-stage reranker after dense, sparse, or hybrid retrieval, especially when agents need cleaner evidence from transcripts, OCR spans, documents, and other text-heavy payloads.
On Mixpeek, Jina Reranker v3 fits after a broad first-stage search. Retrieve a larger candidate set with embeddings, BM25, or hybrid search, then rerank the candidates before sending compact evidence to an agent. This is particularly useful when a video or document pipeline has many short evidence spans that look similar in embedding space.
Architecture
Transformer cross-encoder text-ranking model based on Qwen3-0.6B. The model scores query and document pairs directly, so inference cost scales with the number of candidate pairs reranked.
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-v3 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
- Second-stage reranking for dense, sparse, and hybrid candidate sets
- Multilingual text-ranking model based on Qwen3-0.6B
- Useful for transcript, OCR, caption, and document span precision
- Compact 596M-parameter footprint compared with larger rerankers
- Works well as the precision stage before agent context assembly
Use Cases on Mixpeek
Benchmarks
| Dataset | Metric | Score | Source |
|---|---|---|---|
| Hugging Face | Pipeline tag | text-ranking | Jina AI, 2026 Model Card |
| Hugging Face | Base model | Qwen3-0.6B | Jina AI, 2026 Model Card |
| Hugging Face | Parameters | 596M | Jina AI, 2026 Model Card |
Performance
Common Pipeline Companions
Explore on Mixpeek
Compare alternatives in this category
Hand-picked tools & platforms compared
Deep-dive technical guide
See how Mixpeek runs models as extractors
Store & search embeddings at scale
Usage-based pricing for pipelines
Compare models, APIs & infrastructure
Specification
Research Paper
Jina Reranker v3
arxiv.orgBuild a pipeline with jina-reranker-v3
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