Cohere Launches Command A Reasoning: A Game-Changer for Enterprise Customer Service

Cohere, the Canadian AI startup co-founded by former Transformer paper author Aidan Gomez, has released its first reasoning large language model (LLM), Command A Reasoning. This model is specifically designed for enterprise use, focusing on tasks like customer service, market research, scheduling, and data analysis in secure environments.

Credit: VentureBeat made with Midjourney

Tuned for Enterprises

Command A Reasoning supports up to 256,000 tokens on multi-GPU setups, making it comparable to models like OpenAI's GPT-5. It is a text-only model with 111 billion parameters, trained for tool use and multilingual performance. It supports 23 languages out of the box, including English, French, Spanish, Japanese, Arabic, and Hindi, which is crucial for global enterprises.

Controlled Thinking and Performance

The model features a token budget system, allowing users to toggle reasoning on or off for flexibility in latency and cost. Benchmarks show it outperforms peers like DeepSeek-R1 0528, gpt-oss-120b, and Mistral Magistral Medium in enterprise tasks. It excels in multilingual benchmarks and is designed for conversational tool use, enabling it to call APIs and integrate with existing systems.

Safety and Deployment

Safety is a key focus, with training to avoid over-refusal while filtering harmful content in categories like child safety, self-harm, and violence. The model is available on the Cohere platform and Hugging Face for research under a CC-BY-NC license. Enterprises can contact Cohere for bespoke pricing and private deployments.

Early Adoption

SAP SE is an early adopter, integrating the model to enhance generative AI capabilities in their Business Technology Platform. This collaboration aims to provide customizable agentic applications for enterprise needs.