Don’t sleep on Cohere: Command A Reasoning, its first reasoning model, is built for enterprise customer service and more

Cohere, a Canadian startup co-founded by Aidan Gomez, a notable figure in AI research, has launched its first reasoning-focused large language model (LLM), named Command A Reasoning. This innovative model is designed to cater to various enterprise needs, including customer service, market research, scheduling, and data analysis, all within secure environments. Command A Reasoning boasts impressive technical specifications and early performance benchmarks, highlighting its flexibility, efficiency, and robust reasoning capabilities. Although it operates as a text-only model, it can seamlessly integrate with multimodal systems, enhancing its utility for businesses. For researchers, the model is accessible for non-commercial use, while enterprises must engage with Cohere directly for access, as the company customizes deployments based on individual needs. Recently, Cohere was valued at $6.8 billion after raising $500 million in funding, underscoring its growing significance in the AI sector. Tailored for organizations with extensive document archives and complex workflows, Command A Reasoning can process up to 256,000 tokens on multi-GPU setups, rivaling offerings like OpenAI’s GPT-5. This model incorporates 111 billion parameters and supports 23 languages, including English, French, Spanish, Japanese, Arabic, and Hindi, making it an ideal fit for global enterprises that require consistent quality across markets. Cohere's new platform, North, allows businesses to deploy AI agents and automations on-premises, ensuring complete control over data management while benefiting from advanced reasoning capabilities. The model is particularly adept at automating common enterprise functions such as onboarding and market analysis, designed to enhance efficiency in daily operations. Command A Reasoning also introduces a unique token budget feature, which allows users to allocate reasoning resources based on the task at hand. This flexibility facilitates quicker responses for simpler queries while enabling deeper, more accurate reasoning for complex inquiries. Cohere’s internal benchmarks have shown that Command A Reasoning consistently outperforms competitors like Deep Seek-R1, gpt-oss-120b, and Mistral Magistral Medium on various enterprise reasoning tasks. Its performance in multilingual benchmarks is equally robust, crucial for companies operating in diverse linguistic markets. In addition to its technical prowess, the model has been developed with safety in mind, trained to minimize over-refusal scenarios, and effectively filter harmful content. This makes it particularly suitable for organizations in regulated industries. SAP SE has already partnered with Cohere to integrate this model into its Business Technology Platform, expanding its generative AI capabilities for customized enterprise applications. Command A Reasoning is now available through the Cohere platform and can also be accessed for research purposes on Hugging Face, which provides open weights under a CC-BY-NC license. For enterprises, the value proposition is clear: a versatile model that offers multiple deployment options, fine-tuned performance control, multilingual support, and proven capabilities that indicate superior performance compared to its peers.