| title | Command A |
|---|---|
| subtitle | Our largest, most performant model, ideal for building enterprise agents with a low compute footprint. |
| slug | docs/command-a |
| hidden | false |
| description | Command A is a performant mode good at tool use, RAG, agents, and multilingual use cases. It has 111 billion parameters and a 256k context length. |
| image | ../../../assets/images/edb3e49-cohere_meta_image.jpg |
| keywords | generative AI, Cohere, large language models |
| createdAt | Thu Mar 13 2025 |
| updatedAt | |
| layout | overview |
import { ModelShowcase, Capability, Endpoint} from "../../../components/model-showcase"
<ModelShowcase model={{ name: 'Command A', id: 'command-a-03-2025', capabilities: [ Capability.SafetyModes, Capability.Citations, Capability.ToolUse, Capability.StructuredOutputs, Capability.Multilingual, ], pricing: { input: 2.50, output: 10.0 }, specs: { contextWindow: '256,000', maxOutputTokens: '8,000', knowledgeCutoff: 'June 1, 2024', customSpecs: [ { name: "Hardware", value: "Requires two GPUs to run (A100s / H100s)" } ] }, endpoints: [ Endpoint.ChatV2, Endpoint.ChatV1, Endpoint.ChatCompletions, ], }}/>
Command A is Cohere's most performant model to date, excelling at real world enterprise tasks including tool use, retrieval augmented generation (RAG), agents, and multilingual use cases. At 111B parameters, Command A has a context length of 256K and only requires two GPUs (A100s / H100s) to run, while being significantly more efficient at inference time with 150% higher throughput compared to its predecessor, Command R+ 08-2024.
Command A is excellent for:
- Tool use - With tool use, Command models can be given tools such as search engines, APIs, vector databases, etc., which can expand their baseline functionality. Command A excels at tool use, exhibiting particular strength in using tools in real-world, diverse, and dynamic environments. In addition, Command A is good at avoiding unnecessarily calling tools, which is an important aspect of tool-use in practical applications.
- Agents - As this is being written, agents are among the most exciting frontiers for large language models. Command A’s multistep tool use capabilities allow it to power fast and capable REACT agents. When set up as an internet-augmented research agent, for example, Command A ably completes tasks that require breaking down complex questions into subgoals, and also performs favorably in domains that utilize complex reasoning and active information seeking.
- Retrieval augmented generation - Retrieval Augmented Generation (RAG) refers to the practice of ‘grounding’ model outputs in external data sources, which can increase accuracy. Command A is exceptionally good at generating responses in conversational tasks, attending over long inputs, and extracting and manipulating numerical information in financial settings.
- Multilingual use cases - The model is trained to perform well in 23 languages: English, French, Spanish, Italian, German, Portuguese, Japanese, Korean, Chinese, Arabic, Russian, Polish, Turkish, Vietnamese, Dutch, Czech, Indonesian, Ukrainian, Romanian, Greek, Hindi, Hebrew, Persian. It has been trained to respond in the language of the user, or follow instructions to output a response in a different language. It also excels at performing cross-lingual tasks, such as translation or answering questions about content in other languages.
By default, the model is interactive and optimized for conversation, meaning it is verbose and uses markdown to highlight code. To override this behavior, developers should use a system instruction which asks the model to simply provide the answer and to not use markdown or code block markers. To learn more, consult our documentation on system instructions.