
Independent company · Berlin, Germany
Mixedbread
Applied research lab building retrieval and reranking models, with open embeddings downloaded more than 50 million times.
Berlin, Germany
- Founded
- 2023
- Type
- Independent company
- Based in
- Berlin, Germany
- Key papers
- 1
- Featured projects
- 8
About
Mixedbread is an applied research lab focused on retrieval. It builds embedding, reranking and search systems that give AI applications the right context at the right time.
The team combines deep learning research with high-performance systems engineering and offers its models through a hosted search product while publishing open-source models on Hugging Face. The company states that those open models have been downloaded more than 50 million times.
Recent work covers listwise rerankers, quantization for late-interaction retrieval and ingestion of very large files.
Mission and ambitionsMixedbread aims to build memory for AI, so that intelligent systems have the context they need to understand and act on the world.
Known for
- 01
Open embedding and reranking models
- 02
mxbai-rerank models
- 03
Late-interaction retrieval research
- 04
Hosted multimodal search
Milestones
- 2026
Asymmetric quantization
A June 2026 post describes cutting late-interaction corpus storage 32 times with binary document vectors at near-equal retrieval quality.
- 2026
Dense retrieval analysis
A June 2026 research post argues dense retrievers encode more information than their scoring mechanism expresses.
- 2026
Large-file ingestion system
A July 2026 engineering post describes a slicer and parser architecture for files of any size under bounded compute.
- 2026
mxbai-rerank-v3.1-listwise
A listwise reranker update released in July 2026 inside Mixedbread Search.
Open source
A calm, CLI-native way to semantically grep everything, like code, images, pdfs and more. (4,439 stars)
Baguetter is a flexible, efficient, and hackable search engine library implemented in Python. It's designed fo (214 stars)
The Batched API provides a flexible and efficient way to process multiple requests in a batch, with a primary (161 stars)
Semantic Search for over 50,000 images from the National Gallery of Art collection (66 stars)
Model: 2,194,389 downloads, 847 likes.
Model: 578,018 downloads, 61 likes.
- mixedbread-ai/baguetterdataset
Dataset: 933 downloads, 1 likes.
Model: 541,817 downloads, 59 likes.
Research papers
Notable releases
Openness
Embedding and reranking models are public on Hugging Face and code is mostly Apache-2.0; the hosted product is closed.
People
- Aamir ShakirCo-founder
Co-author of Mixedbread's embedding model papers.
- Julius LippCo-founder (listed)
Listed as co-founder by Crunchbase; current role unclear.
Listed from the lab's own pages and public profiles.
Contact and community
Contact through the website; a support address is printed there.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



