Cohere launches Embed 5 models with multimodal input and cross-tier querying
Cohere has introduced Embed 5, a new embedding model family built for enterprise search, retrieval-augmented generation, and agentic retrieval use cases. The release includes two tiers: Embed 5 Pro, which is aimed at maximum retrieval quality, and Embed 5 Fast, which is tuned for lower latency and lower cost in real-time query workloads. Both models support text, image, and mixed text-image inputs, cover more than 100 languages, and offer a context window of up to 128,000 tokens. Cohere said the two tiers share the same embedding space, allowing users to build indexes with Pro and run queries with Fast to balance quality and efficiency. The models are now available through Cohere API, Model Vault, Microsoft Foundry, and Amazon SageMaker, with support for private VPC or on-premises deployment through vLLM. In benchmark testing, Embed 5 Pro posted an average score of 85.8 on the ViDoRe V3 dataset, ahead of Voyage 4 Large, Gemini Embedding 2, and OpenAI’s text-embedding-3-large. MarkTechPost also said the model ranked first on FinanceBench and other finance-related datasets.