Jina Reranker V2 Base Multilingual
jinaai/jina-reranker-v2-base-multilingual
published Jun 2024 · updated Oct 2025
Jina Reranker V2 Base Multilingual is a transformer-based cross-encoder rerank model fine-tuned for text reranking across multiple languages.
specs
| Task | Reranking (Cross-Encoder) |
| Architecture | Transformer-based cross-encoder |
| License | CC-BY-NC-4.0 (research; commercial via Jina AI APIs, AWS Sagemaker, or Azure Marketplace) |
| Context Length | 1024 tokens (sliding window for longer texts) |
| Task | Reranking (Cross-Encoder) |
| Architecture | Transformer-based cross-encoder |
| License | CC-BY-NC-4.0 (research; commercial via Jina AI APIs, AWS Sagemaker, or Azure Marketplace) |
| Context Length | 1024 tokens (sliding window for longer texts) |
about this model
Jina Reranker v2 (jina-reranker-v2-base-multilingual) is a cross-encoder model fine-tuned for text reranking, scoring the relevance of a query–document pair.
It processes inputs up to 1,024 tokens and uses a sliding window approach to chunk longer texts, reranking each chunk separately. A flash attention mechanism accelerates inference. The model is multilingual and has demonstrated competitiveness across benchmarks for text retrieval, multilingual reranking, function-calling‑aware and text‑to‑SQL‑aware reranking, and code retrieval.
As a hosted API on gigarouter, you can integrate this reranker without managing infrastructure—simply send queries and documents via the OpenAI‑compatible endpoint and receive relevance scores.
best for
- ·Multilingual document reranking in search systems
- ·Retrieval-augmented generation (RAG) pipeline reranking
- ·Code retrieval and function-calling-aware reranking
FAQ
The model supports up to 1024 tokens per input pair; longer texts are handled via a sliding window approach.
The model is licensed under CC-BY-NC-4.0 for research. Commercial use requires a license from Jina AI, AWS Sagemaker, or Azure Marketplace.
Use gigarouter's OpenAI-compatible endpoint with your API key, setting the model to jina-reranker-v2-base-multilingual, and send a rerank request with query and documents.
It is multilingual, supporting languages such as English, German, Spanish, Chinese, Japanese, and more (as shown in the model card examples).
Yes, it uses flash attention by default, offering 3x-6x speedup on compatible GPUs (requires flash-attn and ninja).
The model supports up to 1024 tokens per input pair; longer texts are handled via a sliding window approach.
The model is licensed under CC-BY-NC-4.0 for research. Commercial use requires a license from Jina AI, AWS Sagemaker, or Azure Marketplace.
Use gigarouter's OpenAI-compatible endpoint with your API key, setting the model to jina-reranker-v2-base-multilingual, and send a rerank request with query and documents.
It is multilingual, supporting languages such as English, German, Spanish, Chinese, Japanese, and more (as shown in the model card examples).
Yes, it uses flash attention by default, offering 3x-6x speedup on compatible GPUs (requires flash-attn and ninja).
# rerank documents by relevance; billed per document curl https://gigarouter.ai/v1/rerank \ -H "Authorization: Bearer $GR_KEY" \ -d '{"model":"jinaai/jina-reranker-v2-base-multilingual","query":"capital of France", "documents":["Paris is the capital of France.","Bananas are yellow."]}'
try it live
runs the real hosted model on a shared demo allowance · get your own key + $25 free →