embed-english-v3.0

embed-english-v3.0 is a specialized text embedding model from Cohere designed specifically for English language inputs. It transforms text phrases, sentences, and paragraphs into 1,024-dimensional dense vectors that capture semantic meaning, enabling applications to perform similarity searches, retrieve relevant documents, classify text, and cluster similar items.

The model was trained on approximately 1 billion English language pairs and achieves state-of-the-art performance on widely-used benchmarks including the Massive Text Embedding Benchmark (MTEB) across 56 datasets for retrieval, classification, and clustering, as well as the BEIR benchmark for zero-shot dense retrieval tasks. These benchmark results demonstrate strong generalization across diverse domains.

API access supports batch processing of up to 96 texts per request, with each input limited to 512 tokens maximum. The model requires users to specify an input_type parameter that distinguishes between search_document, search_query, classification, clustering, and image inputs to optimize embeddings for specific use cases. For image inputs, both text and image can be embedded together. Output embeddings are available in multiple formats including float, int8, uint8, binary, ubinary, and base64 representations to optimize for different storage and retrieval requirements.

The model is available through Cohere’s API with usage-based pricing of $0.10 per million tokens, with a free tier available for learning and prototyping subject to rate limits. Deployment options include Cohere’s managed API platform, AWS SageMaker, Azure AI Foundry, and OCI Generative AI service.