Keimodel
Log inStart free
Cohere

Cohere

Enterprise RAG and multilingual AI

Cohere builds language models purpose-built for enterprise deployments. Their Command family excels at retrieval-augmented generation, grounding responses in real documents with cited sources. Founded in 2019 by former Google Brain researchers, Cohere's models are optimised for production reliability, long-context document analysis, and multilingual coverage across dozens of languages through the Aya family.

Founded 2019·4 models·Visit Cohere

4 Models

Command A

Cohere's most capable enterprise model

1 cr
Command A is Cohere's flagship model, designed for complex agentic and RAG workflows. It supports a 256k-token context window, excels at multi-step tool use, and is optimised for enterprise deployments where accuracy and citation quality are critical.
256k ctxTool useJSON mode

Command R+

High-accuracy RAG with grounded citations

3 cr
Command R+ is Cohere's high-performance model for retrieval-augmented generation. It is specifically trained to reason over retrieved documents, produce cited answers, and call tools reliably, making it the go-to model for enterprise search and knowledge management applications.
128k ctxTool useCitations

Command R

Cost-efficient RAG for high-volume pipelines

2 cr
Command R is Cohere's balanced model for production RAG pipelines. It delivers strong citation quality and tool-use capabilities at a fraction of the cost of Command R+, making it ideal for high-volume document question-answering and summarisation workflows.
128k ctxTool useCitations

Aya Expanse 32B

Frontier multilingual model covering 35 languages

1 cr
Aya Expanse 32B is Cohere's multilingual flagship, trained to serve 35 languages with native-level fluency. It significantly outperforms other open-weight models on non-English tasks and is ideal for global applications that require high-quality responses across diverse language markets.
128k ctxMultilingual (35 languages)Tool use

Other providers

OpenAIOpenAIAnthropicAnthropicGoogleGoogleMetaMeta