Command A

Command A is an open weights language model from Cohere and Cohere Labs for researchers and developers building document assistants, chatbots and AI agents. It handles text input and output, with a focus on multilingual business tasks.

• Answers with citations: The model can answer from supplied document snippets and link individual claims to the passages that support them.
• Tool calling: Applications can give it access to APIs, databases and search engines, and its answers can cite the tool results it uses.
• Multilingual text tasks: It supports summarization, information extraction, translation and categorization in languages including English, Arabic, Chinese, Japanese, Ukrainian and Persian.
• Coding assistance: It can generate SQL, explain code and translate code between programming languages.
• Safety modes: The default contextual mode allows broader discussion while rejecting harmful or illegal suggestions; strict mode aims to avoid sensitive topics and profanity.

The weights use Safetensors and work with Hugging Face Transformers for self-hosted inference. Cohere describes deployment on two GPUs. Its long context accommodates substantial amounts of text, though citation-based answers use supplied document snippets. Hosted chat is also available through a playground and Hugging Face Space.

The research weights carry the noncommercial CC-BY-NC-4.0 license and are subject to Cohere Labs' Acceptable Use Policy. Download access requires a Hugging Face account, acceptance of the terms and sharing contact information with Cohere.