mxbai-rerank 0.1.4

Crispy reranking models from Mixedbread.

Mixedbread Reranking Models

Crispy reranking models from Mixedbread. State-of-the-art models for search relevance, powered by reinforcement learning.

Features

• State-of-the-art performance - Outperforms leading open and closed-source rerankers on major benchmarks
• 100+ languages - Strong multilingual support out of the box
• Long context - Handle up to 8k tokens (32k-compatible)
• Code & SQL - Excellent at ranking code snippets and technical content
• Function Call Ranking - Supports reranking of function calls for multi-tool agents
• Fast inference - 8x faster than comparable models
• Easy integration - Drop-in improvement for existing search systems
• Open source - Apache 2.0-licensed, easy to customize
• Managed API - For production use with additional features. We support embeddings, reranking, and an end-to-end multi-modal retrieval solution.

Installation

pip install -U mxbai-rerank

Quick Start

from mxbai_rerank import MxbaiRerankV2

# Initialize the reranker
reranker = MxbaiRerankV2("mixedbread-ai/mxbai-rerank-base-v2")  # or large-v2

# Example query and documents
query = "Who wrote 'To Kill a Mockingbird'?"
documents = [
    "'To Kill a Mockingbird' is a novel by Harper Lee published in 1960.",
    "The novel 'Moby-Dick' was written by Herman Melville.",
    "Harper Lee was born in 1926 in Monroeville, Alabama."
]

results = reranker.rank(query=query, documents=documents)

print(results)

Models

We offer multiple model variants. For more details, see our mxbai-rerank-v2 technical blog post.

• mxbai-rerank-base-v2 (0.5B) - Best balance of speed and accuracy
• mxbai-rerank-large-v2 (1.5B) - Highest accuracy, still with excellent speed

Legacy Models

For more details, see our mxbai-rerank-v1 technical blog post.

• mxbai-rerank-xsmall-v1 (0.1B) - Fastest inference, lower accuracy
• mxbai-rerank-base-v1 (0.2B) - Smaller, faster model
• mxbai-rerank-large-v1 (1.5B) - Large model with highest accuracy

Performance

Benchmark Results

Model	BEIR Avg	Multilingual	Chinese	Code Search	Latency (s)
mxbai-rerank-large-v2	57.49	29.79	84.16	32.05	0.89
mxbai-rerank-base-v2	55.57	28.56	83.70	31.73	0.67
mxbai-rerank-large-v1	49.32	21.88	72.53	30.72	2.24

*Latency measured on A100 GPU

Advanced Usage

Flash Attention Support

The v2 models automatically use Flash Attention 2 when available for faster inference:

pip install flash-attn --no-build-isolation

Long Context Support

reranker = MxbaiRerankV2(
    "mixedbread-ai/mxbai-rerank-base-v2",
    max_length=8192  # Default, can be adjusted up to model limits (32k for v2 models)
)

Instruction Support

results = reranker.rank(query=query, documents=documents, instruction="Figure out the best code snippet for the user query.")

API Access

For managed API access with additional features, such as object reranking and instructions:

from mixedbread import Mixedbread

mxbai = Mixedbread(api_key="YOUR_API_KEY")

results = mxbai.rerank(
    model="mixedbread-ai/mxbai-rerank-large-v2",
    query="your query",
    input=["doc1", "doc2", "doc3"]
)

Training Details

The models were trained using a three-step process:

1. GRPO (Guided Reinforcement Prompt Optimization)
2. Contrastive Learning
3. Preference Learning

For more details, check our technical blog post.

Paper following soon.

Citation

If you use this work, please cite:

@online{v2rerank2025mxbai,
  title={Baked-in Brilliance: Reranking Meets RL with mxbai-rerank-v2},
  author={Sean Lee and Rui Huang and Aamir Shakir and Julius Lipp},
  year={2024},
  url={https://www.mixedbread.com/blog/mxbai-rerank-v2},
}

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a pull request or report an issue on GitHub.

Community & Support

• Discord Community
• X/Twitter
• LinkedIn