Cohere releases Command R

Cohere released Command R, a 35-billion-parameter language model built specifically for enterprise use cases rather than general chat: retrieval-augmented generation, multi-step tool use and long-document processing. It shipped with a 128,000-token context window, well beyond most competing models available at the time, and Cohere said it prioritised reliability and low latency in production deployments over raw benchmark scores.

The release reflected Cohere’s positioning through 2023 and 2024 as a supplier to businesses building on their own data rather than a consumer-chatbot company. Command R was marketed around “RAG at production scale” — grounding model outputs in retrieved documents and citing sources, a capability enterprise customers had flagged as necessary for deployments where unverified hallucination was a liability. The model also supported ten languages and was designed to run efficiently enough for cost-sensitive production use.

Command R was the first of a two-part release: Cohere followed it three weeks later with the larger Command R+, pitched as competitive with GPT-4-class models on enterprise tasks while Command R remained the lighter, cheaper option in the pair. Together they marked Cohere’s clearest attempt to differentiate from OpenAI and Anthropic by competing on retrieval and tool-use workflows rather than general-purpose chat quality.