Embeddings
What are Embeddings?
Embeddings are numerical representations of data (words, phrases, images, etc.) mapped into continuous vector spaces. These vectors capture semantic and contextual relationships, enabling machine learning models to interpret the meaning and similarity between data points mathematically.
Generate your first embeddings
Try out the example to understand how to generate embeddings with our models.
Available Embedding Models
Choose the right model for your use case:
| Model Card | Description |
|---|---|
| mxbai-embed-large-v1 | State-of-the-art English embedding model for semantic search and retrieval. Trained on 700M+ pairs, supports binary embeddings for fast, efficient storage. |
| deepset-mxbai-embed-de-large-v1 | German/English model fine-tuned on 30M+ German pairs. Supports binary quantization and Matryoshka learning for cost-effective, real-world use. |
| mxbai-embed-2d-large-v1 | First 2D-Matryoshka embedding model for flexible speed, storage, and performance trade-offs. Remains competitive even at reduced sizes. |
Check out the API Reference for more information.
Data Models
Understand the core data structures in Mixedbread Vector Stores - Vector Store Files and Chunks - and how they relate to your workflows.
Reranking
Leverage Mixedbread's Reranking API to access state-of-the-art models that re-order search results or candidate lists based on deep semantic relevance. Improve the precision of search, RAG, and recommendation systems.
Last updated: July 14, 2025