Jina Embedding B English V1

jina-embedding-b-en-v1 is a text embedding model that converts textual inputs into dense vector representations, capturing semantic meaning for tasks such as information retrieval and semantic textual similarity. With 110 million parameters and a 768-dimensional output, it offers a balance of inference speed and accuracy. The model was trained on Jina AI’s Linnaeus-Clean dataset, which comprises 380 million curated query-document pairs. As described in the accompanying paper, Jina Embeddings excel in dense retrieval and semantic similarity, and were evaluated using the Massive Text Embedding Benchmark (MTEB).

Benchmark Performance

The table below compares jina-embedding-b-en-v1 against several popular embedding models across standard evaluation tasks (Spearman correlation, except TRECOVID and SciFact which use nDCG and accuracy respectively).

Model	STS12	STS13	STS14	STS15	STS16	STS17	TRECOVID	Quora	SciFact
all-minilm-l6-v2	0.724	0.806	0.756	0.854	0.790	0.876	0.473	0.876	0.645
all-mpnet-base-v2	0.726	0.835	0.780	0.857	0.800	0.906	0.513	0.875	0.656
ada-embedding-002	0.698	0.833	0.761	0.861	0.860	0.903	0.685	0.876	0.726
jina-embedding-b-en-v1	0.751	0.809	0.761	0.856	0.812	0.890	0.606	0.876	0.594

The model achieves top performance on STS12 among the compared models and delivers competitive results across other semantic similarity and retrieval tasks. For further details on training data, methodology, and a negation-aware dataset, refer to the Jina Embeddings paper (arXiv:2307.11224).