Cohere Unveils Command A Reasoning To Power Enterprise Applications With Advanced Performance, Efficiency, And Control

In Brief

Cohere released Command A Reasoning, a new enterprise reasoning model that outperforms similar rivals like gpt-oss and DeepSeek R1 on agentic benchmarks.

Cohere, a company specializing in natural language processing and large language models (LLMs), announced that it has launched Command A Reasoning, its most advanced model designed for enterprise reasoning tasks. The model demonstrates superior performance compared with other privately deployable models in its class, including gpt-oss-120b, DeepSeek-R1 0528, and Mistral Magistral Medium.

Command A Reasoning is tailored for enterprise deployment, offering secure, efficient, and scalable options. For smaller setups, it can operate on a single H100 or A100 GPU with a context length of 128k, while multi-GPU deployments optimized for latency can scale to a 256k context length. This flexibility allows organizations to maximize hardware utilization. Its extended context capabilities make it particularly suitable for document-intensive workflows and complex multi-step agentic applications.

Organizations can set token budgets to manage compute usage and control costs, enabling the model to handle tasks requiring maximum accuracy or be configured for faster, high-throughput operations. Command A Reasoning serves as the core generative model for North, Cohere’s secure agentic AI platform, allowing enterprises to deploy customized AI agents and automations on-premises, leveraging the model’s advanced reasoning capabilities.

Enterprise Reasoning Model Demonstrates Strong Performance Across Key Business Tasks

Command A Reasoning delivers top-tier performance across major agentic benchmarks and supports a wide range of business languages, enabling global enterprises to deploy agents with consistent quality. The model is particularly effective in powering end-to-end systems involving chained or hierarchical agents and in selecting the most relevant tools to complete tasks. A notable application is the Deep Research system, which surpasses similar capabilities from other leading AI labs.

Deep Research is designed to handle complex, in-depth questions that require more than a simple search. It generates detailed, well-sourced reports in minutes—tasks that would typically take employees hours—by using a multi-agent architecture powered by Command A Reasoning. The system breaks down user requests into smaller research topics, then multiple AI agents search and analyze information from diverse sources in parallel, before consolidating verified findings into a single structured report directly addressing the original query.

The model enables scalable business applications across industries, combining high accuracy with efficient compute usage. Its low hardware requirements allow private deployments on a single H100 or A100 GPU. Users can manage compute and costs through a token budget, dynamically balancing precision for mission-critical tasks with high-throughput operations, eliminating the need for separate reasoning and non-reasoning models and maximizing GPU efficiency. North’s internal tests show smooth performance scaling from efficient responses at minimal reasoning to deeper, more detailed outputs at higher reasoning levels, with Command A Reasoning outperforming its predecessor even at zero reasoning.

Safety is a core principle in the training and evaluation of Command A Reasoning. The model is designed to balance avoiding over-refusal of valid requests while preventing the dissemination of harmful or malicious content. Focus areas include child sexual exploitation and abuse, self-harm, violence and hate, sexual content, and conspiracy theories.

Command A Reasoning is currently available on the Cohere platform and for research use via Hugging Face.

Cohere develops foundation models and end-to-end AI solutions aimed at addressing practical business challenges. Recently, the company partnered with Ensemble Health Partners to integrate secure agentic AI into complex healthcare workflows, providing providers, clinical staff, and revenue cycle teams with an advanced AI system designed to reduce administrative workloads while maintaining the high security and compliance standards required in healthcare operations.