Cohere, a Canadian AI startup, has unveiled Command A, its latest generative AI model designed specifically for enterprise applications. Command A succeeds the company's previous models, Command R and Command R+, and is tailored to enhance retrieval-augmented generation (RAG), tool use, and enterprise AI efficiency. The new model boasts a 256,000 token context window, doubling the capacity of its predecessor, and supports 23 languages, including improved handling of Arabic dialects. Command A is designed to match or outperform competitors such as OpenAI's GPT-4o and DeepSeek-V3 in business, STEM, and coding tasks, while requiring significantly less computational power. It features 111 billion parameters and achieves a token generation rate of 156 tokens per second. Notably, Command A can operate on just two GPUs, making it an efficient choice for businesses looking to deploy AI solutions with lower hardware requirements. It achieves faster token generation rates and reduced latency compared to its rivals, offering a cost-effective and high-performance solution for enterprises. Private deployments are available and can be up to 50% cheaper than API-based access. Cohere's focus on building models for business customers, rather than pursuing artificial general intelligence, has allowed the company to achieve these results with a training cost of less than US$30 million. Command A is now available on the Cohere platform and with open weights for research use on Hugging Face. The pricing is set at $2.50 per million input tokens and $10.00 per million output tokens. In benchmark evaluations, Command A scored 98.2% accuracy in responding to English prompts in Arabic and achieved an ADI2 score of 24.7 for dialect consistency.