mxbai-embed
Mixedbread’s mxbai-embed models produce text vectors locally for document retrieval, with Apache-2.0 weights and adjustable embedding dimensions.
mxbai-embed is Mixedbread’s downloadable text embedding model family for developers building their own retrieval systems. It converts queries and passages into vectors that an application can compare for relevance or similarity.
The mxbai-embed-large-v1 model supports local inference with SentenceTransformers, Transformers and Transformers.js. Its official examples encode document batches and compare them with a query. Retrieval queries use the documented search prefix; documents do not need that prefix.
The model supports Matryoshka Representation Learning, letting developers shorten embeddings to reduce storage. The examples also show binary quantization of embedding vectors. These are changes to output representations, distinct from quantizing model weights.
The large-v1 weights are licensed under Apache-2.0 and target English text. The model card includes an Infinity Docker serving example and a hosted API option, but local inference does not require the hosted service. Mixedbread’s managed search platform and its newer platform-only models have separate capabilities and deployment terms.
