Command A
Cohere's flagship enterprise model, tuned for agentic and RAG workloads
Command A is Cohere’s flagship large language model, released in March 2025 at the top of its Command lineup. It has 111 billion parameters and a 256,000 token context window, and Cohere built it to run on as few as two GPUs while matching or beating larger competing models on agentic, multilingual, and coding tasks. It scores 85.5 on MMLU, putting it in the same range as GPT-4o and Llama 3.3 70B. Command A is tuned for retrieval-augmented generation and tool use, and it’s available through the Cohere platform and Azure AI at $2.50 per million input tokens and $10 per million output tokens, or as a self-hosted deployment for enterprises that need to keep data on their own infrastructure. Cohere later added Command A+, a larger 218 billion parameter model, to the same family.