News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
fnCommand A

Command A

Cohere's enterprise-focused large language model for agents, search and tool use

Command A is a large language model built by Cohere for real-world business work, especially agents, retrieval-augmented generation and tool use. A technical report was posted in April 2025, and the company says it matches or beats GPT-4o and DeepSeek-V3 on enterprise agentic tasks while using less compute.

Share on WhatsApp

  • Agent-optimised model supporting 23 languages of global business
  • Cohere claims parity or better than GPT-4o and DeepSeek-V3 on enterprise agentic tasks
  • Hybrid architecture designed to balance efficiency with performance
  • Training used self-refinement algorithms and model merging
  • Weights released for research purposes, alongside Command R7B

What it is

Command A is a large language model from Cohere, aimed at enterprise work rather than general consumer chat. The company describes it as agent-optimised and multilingual, covering 23 languages used in global business. It is built for retrieval-augmented generation, where a model pulls in outside documents to ground its answers, and for tool use, where it calls other software to carry out steps in a business process. A technical report on its development appeared in April 2025. Cohere positions it as delivering strong performance on real business tasks while needing comparatively little compute to run.

How it came about

Cohere published a technical report on Command A on the arXiv preprint server on 1 April 2025, with a revised version on 14 April. The report was written by a large Cohere team, with 228 authors listed. Alongside Command A it also covers Command R7B, a smaller model sharing capability and architectural similarities. The company's own blog announced the model as offering performance on a par with or better than GPT-4o and DeepSeek-V3 on agentic enterprise tasks.

Inside the model

Cohere describes Command A as using a novel hybrid architecture that balances efficiency against high performance. Training took a decentralised approach, combining self-refinement algorithms with model merging techniques, where separate model versions are blended. The model is optimised for retrieval-augmented generation with grounding, and for tool use in automated business workflows. It supports 23 languages of global business. The technical report also presents Command R7B, which shares architectural and capability traits with Command A. Weights for both were released for research purposes.

What it can do

Cohere reports that Command A performs on a par with or better than GPT-4o and DeepSeek-V3 across agentic enterprise tasks, with notably greater efficiency. The company highlights best-in-class retrieval-augmented generation with grounding, plus tool use for automating complex business processes. Its language coverage spans 23 languages used in global business. The technical report presents evaluations across enterprise-relevant tasks and public benchmarks, though specific scores are not given here.

How to use it

Weights for Command A were released for research purposes, alongside those of Command R7B. The model is also offered by Cohere as part of its enterprise AI platform, which includes dedicated model inference and search products.

Questions people ask about Command A

What is Command A for?

It is built for enterprise work: retrieval-augmented generation with grounding, tool use to automate business processes, and agentic tasks. Cohere says it handles 23 languages used in global business and runs with comparatively little compute.

Who made it and when did it appear?

Cohere built it. A technical report was posted on arXiv on 1 April 2025, revised on 14 April 2025, written by a team of 228 authors. Cohere's blog announced the model alongside claims about its efficiency.

Can I download the model?

Weights for Command A were released for research purposes. The technical report also covers Command R7B, a smaller related model, whose weights were released the same way.

How does it compare to other models?

Cohere states that Command A is on a par with or better than GPT-4o and DeepSeek-V3 on agentic enterprise tasks, while being significantly more efficient. These are the company's own claims from its blog and report.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Command A yet, so the facts on this page come from the sources listed above: cohere.com/blog/command-a, arxiv.org/abs/2504.00698. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.