Cohere AI launches Command A: high-performance, cost-efficient enterprise AI model

Cohere (AI) has released its latest enterprise AI model, Command A, designed to deliver high performance with minimal compute requirements. Command A competes with models like GPT-4o and DeepSeek-V3, particularly excelling in business, STEM, and coding tasks, while offering greater efficiency. It operates on just two GPUs (A100 or H100), compared to the 32 GPUs needed by some competitors, significantly reducing infrastructure costs.

The model processes 156 tokens per second, making it 1.75 times faster than GPT-4o and 2.4 times faster than DeepSeek-V3, with a reduced latency of 6,500ms for time-to-first-token. It supports a context length of 256,000 tokens, double that of its predecessor, Command-R, allowing for larger inputs. Command A also offers multilingual capabilities, supporting 23 global languages with enhanced Arabic dialect matching.

Command A is integrated with Cohere’s North AI platform, supporting retrieval-augmented generation (RAG) and agentic tool use, aligning with an enterprise-first strategy. It provides cost-effective private deployments, up to 50% cheaper than API-based access. The model is available on Cohere’s platform and for research on Hugging Face under a CC-BY-NC 4.0 license, with pricing set at $2.50 per million input tokens and $10.00 per million output tokens.