Jina Reranker V1 Turbo En
jinaai/jina-reranker-v1-turbo-en
published Apr 2024 · updated Sep 2025
Jina Reranker V1 Turbo En is a rerank model that uses knowledge distillation from a larger teacher model to deliver fast and competitive reranking of documents against a query.
specs
| Task | Reranking |
| Architecture | JinaBERT with symmetric bidirectional ALiBi |
| Parameters | 37.8 million |
| License | Apache-2.0 |
about this model
jina-reranker-v1-turbo-en is a reranking model that delivers fast, high-quality document ranking by leveraging knowledge distillation from a larger teacher model. Built on JinaBERT — a BERT variant with a symmetric bidirectional implementation of ALiBi (Attention with Linear Biases) — it can process sequences up to 8,192 tokens, enabling effective reranking of long documents without truncation. The model uses 6 transformer layers and 37.8 million parameters, striking a balance between speed and accuracy.
Model Architecture Comparison
| Model | Layers | Hidden Size | Parameters (M) |
|---|---|---|---|
| jina-reranker-v1-base-en | 12 | 768 | 137.0 |
| jina-reranker-v1-turbo-en | 6 | 384 | 37.8 |
| jina-reranker-v1-tiny-en | 4 | 384 | 33.0 |
Benchmark Performance
Evaluated on three key benchmarks, the model achieves competitive results:
| Model | NDCG@10 (17 BEIR datasets) | NDCG@10 (5 LoCo datasets) | Hit Rate (LlamaIndex RAG) |
|---|---|---|---|
| jina-reranker-v1-base-en | 52.45 | 87.31 | 85.53 |
| jina-reranker-v1-turbo-en | 49.60 | 69.21 | 85.13 |
| jina-reranker-v1-tiny-en | 48.54 | 70.29 | 85.00 |
| mxbai-rerank-base-v1 | 49.19 | – | 82.50 |
| mxbai-rerank-xsmall-v1 | 48.80 | – | 83.69 |
| ms-marco-MiniLM-L-6-v2 | 48.64 | – | 82.63 |
| ms-marco-MiniLM-L-4-v2 | 47.81 | – | 83.82 |
| bge-reranker-base | 47.89 | – | 83.03 |
On the 17 BEIR datasets, jina-reranker-v1-turbo-en ranks second overall among all compared models. The LoCo dataset results are not available for other models because they do not support documents longer than 512 tokens.
The model is available under the Apache-2.0 license and has accumulated over 1.1 million downloads on Hugging Face.
best for
- ·Reranking search results for long documents up to 8,192 tokens
- ·Building fast RAG pipelines where low latency is critical
- ·Improving relevance in e-commerce product search
FAQ
The model supports up to 8,192 tokens per query-document pair.
It is faster (6 layers, 37.8M parameters) than the base version (12 layers, 137M parameters) while achieving an NDCG@10 of 49.60 on 17 BEIR datasets versus 52.45 for the base.
It is released under the Apache-2.0 license.
Use the OpenAI-compatible endpoint with your API key, specifying the model name jina-reranker-v1-turbo-en in the request.
It expects a query and a list of documents; each document is scored for relevance to the query.
We're benchmarking and onboarding Jina Reranker V1 Turbo En as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.