Cohere announced Command A on March 13, 2025 and described the model as a launch aimed at enterprise agentic tasks with lower compute requirements.
Cohere's documentation says Command A has 111 billion parameters, a 256k context length and strengths in tool use, retrieval augmented generation, agents and multilingual use cases.
The Command A documentation says the model requires two A100 or H100 GPUs to run and delivers 150% higher inference throughput than its predecessor, Command R+ 08-2024.
Cohere's launch post says Command A can deliver up to 156 tokens per second and says private deployments can be up to 50% cheaper than API-based access.
Cohere says Command A supports 23 languages, including English, French, Spanish, Italian, German, Portuguese, Japanese, Korean, Chinese, Arabic, Russian, Polish, Turkish, Vietnamese, Dutch, Czech, Indonesian, Ukrainian, Romanian, Greek, Hindi, Hebrew and Persian.
Cohere's model overview says the Command family powers tool-using agents, retrieval augmented generation, translation, copywriting and similar use cases through the Chat endpoint.
Cohere says Command A's tool-use path can connect models to search engines, APIs and vector databases, and it says the model is built for real-world, diverse and dynamic tool environments.
For governance teams, the concrete announcement is a model release tied to private deployment, long-context enterprise data use, tool use and RAG; the cited Cohere pages do not provide an independent audit of the benchmark or cost claims.
