Multilingual Reranker (base, 308M)

A cross-encoder reranker for search and RAG in 40+ languages, built on mmBERT-base. Give it a query and candidate passages, for example the top 20-100 hits of a vector or keyword search. It scores each pair, so you can re-sort the candidates and keep the best ones for your LLM. Queries and passages can be in different languages. It is distilled from Qwen3-Reranker-4B into a model about 13x smaller. Apache-2.0, trained on openly licensed web text. ONNX files for CPU and the browser (transformers.js) are included. A smaller, faster version is available as Horizon-Labs/multilingual-reranker-small. Try it in the browser.

• One output logit per (query, passage) pair: higher = more relevant. sigmoid(logit) gives a 0-1 relevance score.
• Max length: trained with 384 tokens per pair. Longer passages are truncated; split long documents into chunks.
• onnx/model_quantized.onnx (int8 embeddings, 641 MB): scores differ from fp32 by 0.001 on average (sigmoid scale) over 1240 benchmark pairs.

Usage

sentence-transformers:

from sentence_transformers import CrossEncoder

model = CrossEncoder("Horizon-Labs/multilingual-reranker-base")
query = "How tall is the Eiffel Tower?"
passages = ["The Eiffel Tower is 330 metres tall.", "La tour Eiffel a été construite pour l'Exposition universelle de 1889.",
            "The Statue of Liberty is 93 metres tall."]
print(model.rank(query, passages))   # [{'corpus_id': 0, 'score': ...}, ...]

transformers:

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

tok = AutoTokenizer.from_pretrained("Horizon-Labs/multilingual-reranker-base")
model = AutoModelForSequenceClassification.from_pretrained("Horizon-Labs/multilingual-reranker-base").eval()
enc = tok([query] * len(passages), passages, padding=True, truncation=True, max_length=512, return_tensors="pt")
with torch.no_grad():
    scores = model(**enc).logits[:, 0].sigmoid()

transformers.js:

import { AutoTokenizer, AutoModelForSequenceClassification } from "@huggingface/transformers";
const tok = await AutoTokenizer.from_pretrained("Horizon-Labs/multilingual-reranker-base");
const model = await AutoModelForSequenceClassification.from_pretrained("Horizon-Labs/multilingual-reranker-base", { dtype: "q8" });
const enc = await tok(new Array(passages.length).fill(query), { text_pair: passages, padding: true, truncation: true });
const { logits } = await model(enc);

RAG frameworks (checked November 2026 with langchain 1.4, llama-index-core 0.14 and haystack-ai 3.3; all three load the model through sentence-transformers' CrossEncoder):

# LangChain: pip install langchain-community langchain-classic sentence-transformers
from langchain_community.cross_encoders import HuggingFaceCrossEncoder
from langchain_classic.retrievers.document_compressors import CrossEncoderReranker
reranker = CrossEncoderReranker(model=HuggingFaceCrossEncoder(model_name="Horizon-Labs/multilingual-reranker-base"), top_n=3)
top_docs = reranker.compress_documents(documents, query)   # use with ContextualCompressionRetriever

# LlamaIndex: pip install llama-index-core sentence-transformers
from llama_index.core.postprocessor import SentenceTransformerRerank
reranker = SentenceTransformerRerank(model="Horizon-Labs/multilingual-reranker-base", top_n=3)   # node_postprocessors=[reranker]

# Haystack 3: pip install haystack-ai sentence-transformers-haystack
from haystack_integrations.components.rankers.sentence_transformers import SentenceTransformersSimilarityRanker
ranker = SentenceTransformersSimilarityRanker(model="Horizon-Labs/multilingual-reranker-base", top_k=3)

Evaluation

nDCG@10 for reranking the candidate lists of public reranking benchmarks (the MTEB versions). Since v1.1 the training data includes MIRACL's training split (real queries; different queries from the MIRACL dev queries tested here), so the MIRACL column is in-domain for this model, as it is for bge-reranker-v2-m3. The other benchmarks were used only for evaluation. Every model sees the same queries and candidates, through the same script (code/). MIRACL: 60 queries per language with 100 candidates each. Wikipedia (WikipediaRerankingMultilingual): 60 queries per language with 9 candidates each. "Other" = ESCI (es, jp, us), RuBQ, T2Reranking, mMARCO-ja and AskUbuntu.

• The table shows the released checkpoint. Means over 2 training seeds (mean nDCG@10 .854 / .855).
• For reference, ranking the same candidates by bge-m3 dense-embedding similarity alone scores 0.797 / 0.927 / 0.788 (mean 0.837), so on this test a reranker adds most on top of a weaker first-stage retriever (BM25 or a small embedding model). Full comparison: leaderboard.
• Models ahead of this one: MIRACL: bge-reranker-v2-m3, Qwen3-Reranker-4B; Wikipedia: Qwen3-Reranker-4B; other: bge-reranker-v2-m3, Qwen3-Reranker-4B. bge-reranker-v2-m3 and, since v1.1, this model were trained on MIRACL training queries (not the dev queries used here).

Per set (nDCG@10):

Training

• Passages: about 400k passages of 2-6 sentences from FineWeb-2 and FineWeb (ODC-BY) in 47 languages.
• Queries: Qwen3.8-27B (Apache-2.0) wrote a natural question and a keyword query for each passage, in the passage's language, plus English questions for 15% of the non-English passages (cross-lingual search).
• Real queries (since v1.1): 40,203 Wikipedia search queries in 16 languages from the MIRACL training split (Apache-2.0), with their human-judged relevant and non-relevant passages (padded to 16 candidates with random passages of the same language), scored by the same teacher.
• Scale: 334,400 synthetic queries with 16 candidates each (5.35M teacher-scored pairs), sampled evenly across languages (plus the 40,203 MIRACL queries). The synthetic data is published as Horizon-Labs/multilingual-rerank-distill.
• Schedule: 2 epochs, learning rate 3e-5 (2 epochs were chosen over 1 by validation loss, not by the benchmarks) (v1.1: the run stopped at step 44,000 of 46,348, about 1.9 epochs, at the job time limit; that checkpoint was also the best by validation loss), 16 queries x 16 candidates per step, max 384 tokens per pair.
• Candidates: for each query, the 15 most similar passages in the same language by bge-m3 dense retrieval (hard negatives) plus the source passage.
• Labels: Qwen3-Reranker-4B (Apache-2.0) scored every (query, candidate) pair. The student learns the teacher's ranking with a listwise KL loss over each query's 16 candidates, plus a pointwise loss on the teacher's relevance probability. The teacher's scores are soft labels, so the near-duplicate passages that dense retrieval finds are not wrongly treated as negatives.
• Code: code/ in this repository.

Changelog

• v1.1 (2026-10-16): added 40,203 real search queries from the MIRACL training split (16 languages) to the distillation data. 2-seed means v1.0 -> v1.1: MIRACL .760 -> .798 (in-domain since v1.1), Wikipedia .964 -> .967, other .796 -> .797, mean .840 -> .854. v1.0 stays available under the tag v1.0.

Limitations

• The model inherits the teacher's judgement, including its mistakes. It is weaker than the teacher, especially on long or technical passages.
• Training queries are mostly LLM-written questions and keyword queries over web text (plus real Wikipedia search queries since v1.1). Very domain-specific search (legal, medical, code) and conversational queries are less covered.
• Pairs are truncated at the max length. Rerank passage-sized chunks, not whole documents.