MXBAI Reranker Base
mixedbread-ai/mxbai-rerank-base-v1
published Feb 2024 · updated Apr 2025
MXBAI Reranker Base is a rerank model that reorders search results by semantic relevance to improve accuracy over keyword-based retrieval.
specs
| Task | text-ranking |
| Architecture | DeBERTa-v2 |
| License | Apache-2.0 |
about this model
mxbai-rerank-base-v1 is a reranking model that improves the relevance of search results by reordering top candidates from an initial retrieval stage. It is based on the DeBERTa-v2 architecture and is trained on a large corpus of real-world search queries with top-10 results from search engines, where an LLM judges relevance. This approach enables the model to add semantic understanding to existing keyword-based search infrastructure such as Elasticsearch, OpenSearch, or Solr without requiring changes to the retrieval pipeline. The model is English-only and supports the text-ranking pipeline.
The model is part of a family of three sizes: mxbai-rerank-xsmall-v1 (capacity-efficient), mxbai-rerank-base-v1 (balanced performance and size), and mxbai-rerank-large-v1 (highest accuracy). Performance is on par with or exceeds closed-source alternatives on industry-relevant use cases. No specific benchmark numbers were provided in the original sources.
Available formats include ONNX, Safetensors, and Transformers.js. Licensed under Apache-2.0.
best for
- ·Boosting keyword-based search systems (Elasticsearch, OpenSearch, Solr) with a semantic reranking stage
- ·Re-ranking top-10 results from a search engine to improve relevance for user queries
FAQ
It is designed to add semantic relevance as a final reranking step after keyword search, improving result quality without changing existing infrastructure.
The family has three sizes: xsmall (capacity-efficient), base (balance of size and performance), and large (highest accuracy). This is the base variant.
It is released under the Apache-2.0 license, allowing commercial and personal use with attribution.
Input is a query and a list of passages (texts). Output is a list of relevance scores (one per passage), typically confidence values.
Use the gigarouter OpenAI-compatible endpoint with your API key, following the standard chat/completions or rerank-specific schema.
We're benchmarking and onboarding MXBAI Reranker Base 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.