Cohere Launches Command R+ for Enterprise Retrieval at Scale

Cohere has launched Command R+, a 104-billion parameter language model designed specifically for enterprise retrieval-augmented generation (RAG) at scale. Unlike OpenAI and Anthropic models, Cohere deploys Command R+ on-premise or in customer-managed cloud infrastructure — a critical requirement for financial services, healthcare, and government customers.

What Makes It Different

Command R+ was trained with 128K context length and optimised for grounded generation — the ability to cite sources accurately without hallucination. In internal benchmarks, it outperforms GPT-4o on enterprise RAG tasks by 12% while costing 60% less per token at scale. The model is already deployed by three major European banks and two US federal agencies.