Embed 5 Pro

Embed 5 Pro is the larger of two embedding models that Cohere released on 30 September 2026. It turns text, images, combined text and images, and parsed documents into vectors for enterprise search, retrieval-augmented generation and classification. Cohere says it covers more than 100 languages, with particular attention to European languages.

Main features

• Multimodal embeddings: text, images, fused text and images, and parsed documents, each mapped to a vector.
• Long context: accepts up to 128,000 tokens of input.
• Adjustable output: vectors of 2048, 1536, 1024, 768, 512 or 256 dimensions, as float, int8 or binary values, so storage can be traded against accuracy.
• Wide language coverage: includes Japanese, Chinese, Korean, Arabic and Hindi among its languages.

Clients

Enterprise teams building search, retrieval and classification systems. Cohere also lists integrations with the main vector databases and search engines.

Key facts

• Released: 30 September 2026
• Availability: Cohere's API and Model Vault, Microsoft Foundry, Amazon SageMaker, the North platform, and private deployment
• Benchmarks: Cohere's own results on visual-document, financial and parsed-PDF retrieval, with no independent figures found