Ruri V3 Reranker 310M
cl-nagoya/ruri-v3-reranker-310m
published Apr 2025 · updated Apr 2025
Ruri V3 Reranker 310M is a Japanese general-purpose reranker model built on ModernBERT-Ja, supporting up to 8192 tokens and achieving state-of-the-art performance on Japanese text ranking benchmarks.
specs
| Task | Reranking (Text Ranking) |
| Architecture | Cross-Encoder based on ModernBERT-Ja |
| Parameters | 315M |
| Max Sequence Length | 8192 tokens |
| Language | Japanese |
| License | Apache 2.0 |
about this model
Benchmarks
The model achieves state-of-the-art results on Japanese retrieval benchmarks:| Model | Parameters (w/o Emb.) | JQaRA nDCG@10 | JaCWIR MAP@10 | MIRACL Recall@30 |
|---|---|---|---|---|
| Ruri-v3-reranker-310m | 315M (236M) | 86.9 | 95.4 | 97.3 |
| hotchpotch/japanese-reranker-cross-encoder-xsmall-v1 | 107M (11M) | 61.4 | 93.8 | 90.6 |
| hotchpotch/japanese-reranker-cross-encoder-small-v1 | 118M (21M) | 62.5 | 93.9 | 92.2 |
| hotchpotch/japanese-rer |
best for
- ·Re-ranking search results for Japanese queries
- ·Improving retrieval accuracy in Japanese RAG pipelines
- ·Ranking candidate documents for Japanese question answering
FAQ
It is best used for re-ranking search results and improving the quality of retrieved documents in Japanese information retrieval and RAG systems.
According to benchmarks, Ruri V3 Reranker 310M achieves higher nDCG@10, MAP@10, and Recall@30 than earlier Japanese reranker models, making it state-of-the-art.
It expects pairs of query and document text. For example, using a CrossEncoder in sentence-transformers with a list of [query, document] pairs.
Use the gigarouter OpenAI-compatible endpoint with your API key, sending a request with a query and a list of documents to be ranked.
The model is released under the Apache License, Version 2.0.
We're benchmarking and onboarding Ruri V3 Reranker 310M 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.