Cohere Unveils Command A: A Powerful, Efficient AI Model for Enterprise Applications

Cohere targets global enterprises with new highly multilingual Command A model requiring only 2 GPUs

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Canadian AI startup Cohere -- co-founded by one of the authors of the original Transformer paper that kickstarted the large language model (LLM) revolution back in 2017 -- today unveiled Command A, its latest generative AI model designed for enterprise applications. As the successor to Command-R, which debuted in March 2024, and Command R+ following it, Command A builds on Cohere's focus on retrieval-augmented generation (RAG), tool use, and enterprise AI efficiency -- especially with regards to compute and the speed at which it serves up answers. That's going to make it an attractive option for enterprises looking to gain an AI advantage without breaking the bank, and for applications where prompt responses are needed -- such as finance, health, medicine, science, law and more. With faster speeds, lower hardware requirements, and expanded multilingual capabilities, Command A positions itself as a strong alternative to models such as GPT-4o and DeepSeek-V3 -- classic LLMs, not the new reasoning models that have taken the AI industry by storm lately. Unlike its predecessor, which supported a context length of 128,000 tokens (referencing the amount of information the LLM can handle in one input/output exchange, about equivalent to a 300-page novel), Command A doubles the context length to 256,000 tokens (equivalent to 600 pages of text) while improving overall efficiency and enterprise readiness. It also comes on the heels of the release of Cohere for AI -- the non-profit subsidiary of the company -- releasing an open source (for research only) multilingual vision model called Aya Vision earlier this month. A step up from Command-R When Command-R launched in early 2024, it introduced key innovations like optimized RAG performance, better knowledge retrieval, and lower-cost AI deployments. It gained traction with enterprises, integrating into business solutions from companies like Oracle, Notion, Scale AI, Accenture, and McKinsey, though a November 2024 report from Menlo Ventures surveying enterprise adoption put Cohere's market share among enterprises at a slim 3%, far below OpenAI (34%), Anthropic (24%), and even small startups like Mistral (5%). Now, in a bid to become a bigger enterprise draw, Command A pushes these capabilities even further. According to Cohere, it: Strengthens multilingual AI capabilities, with improved Arabic dialect matching and expanded support for 23 global languages Built for the enterprise Cohere has continued its enterprise-first strategy with Command A, ensuring that it integrates seamlessly into business environments. Key features include: Multilingual and highly performant in Arabic A standout feature of Command A is its ability to generate accurate responses across multiple languages, including improved handling of Arabic dialects. In benchmark evaluations: Built for speed and efficiency Speed is a critical factor for enterprise AI deployment, and Command A has been engineered to deliver results faster than many of its competitors. Private and on-prem deployments are available upon request. Industry reactions Several AI researchers and Cohere team members have shared their enthusiasm for Command A. Dwaraknath Ganesan, Pretraining at Cohere, commented on X: "Extremely excited to reveal what we have been working on for the last few months! Command A is amazing. Can be deployed on just 2 H100 GPUs! 256K context length, expanded multilingual support, agentic tool use... very proud of this one." Pierre Richemond, AI Researcher at Cohere, added: "Command A is our new GPT-4o/DeepSeek v3 level, open-weights 111B model sporting a 256K context length that has been optimized for efficiency in enterprise use cases." Building on the foundation of Command-R, Cohere's Command A represents the next step in scalable, cost-efficient enterprise AI. With faster speeds, a larger context window, improved multilingual handling, and lower deployment costs, it offers businesses a powerful alternative to existing AI models.

Cohere releases a low-cost AI model that requires only two GPUs - SiliconANGLE

Cohere releases a low-cost AI model that requires only two GPUs Artificial intelligence startup Cohere Inc. today unveiled Command A, its latest large language model capable of high-performance capabilities for business needs with minimal hardware requirements than competitors' AI models. The startup touted the LLM as capable of exceeding leading proprietary and open models such as OpenAI GPT-4o and DeepSeek-V3. The company added that in private deployments the LLM can run across two graphics processing units, Nvidia Corp.'s A100 or H100, while competing models can take up to 32. This size differential can be important because customers that require internal deployments, such as finance and healthcare, often must place their AI models inside their firewalls. This means buying costly AI accelerator hardware and having high-performing models that can run within their enterprise perimeter is a must. "In head-to-head human evaluation across business, STEM, and coding tasks, Command A matches or outperforms its larger and slower competitors -- while offering superior throughput and increased efficiency," Cohere said. It detailed that Command A can deliver tokens at a rate of up to 156 tokens/sec, which is 1.75x faster than GPT-4o and 2.4x faster than DeepSeek-V3. With business use in mind, the model also has a larger context window at 256,000 tokens, which is twice the size of the industry average, including Cohere's Command R+ model. It means that the model can ingest a sizable number of documents at once or up to a 600-page book. "We are only training our model to make you better at your job," Cohere co-founder Nick Frosst said. "It should feel like getting into a mech for your mind. So, we are training it to empower you. So, it should feel specifically good at that." The company stated that it focused on developing capabilities in the model designed to enable the scalable operation of AI agents. Agentic AI has recently become a prominent trend in the industry, aiming to create artificial intelligence systems that can analyze data, make decisions and carry out tasks with minimal or no human involvement. In practice, this requires massive amounts of compute and doing so efficiently and accurately based on company information requires well-trained AI models. Cohere said Command A will integrate directly with its secure AI agent platform, North, which allows enterprise business users to use the full potential of their company data. The platform is designed to enable enterprise AI agents to use customer relationship management, resource planning software and other tools to automate tasks.

Cohere releases a low-cost AI model that requires only 2 GPUs - SiliconANGLE

Artificial intelligence startup Cohere Inc. today unveiled Command A, its latest large language model capable of high-performance capabilities for business needs with minimal hardware requirements in comparison to competitors. The startup touted the LLM as capable of exceeding leading proprietary and open models such as OpenAI GPT-4o and DeepSeek-V3. The company added that in private deployments the LLM can run across two graphics processing units, Nvidia Corp.'s A100 or H100, while competing models can take up to 32. This size differential can be important because customers that require internal deployments, such as finance and healthcare, often must place their AI models inside their firewalls. This means buying costly AI accelerator hardware and having high-performing models that can run within their enterprise perimeter is a must. "In head-to-head human evaluation across business, STEM, and coding tasks, Command A matches or outperforms its larger and slower competitors -- while offering superior throughput and increased efficiency," Cohere said. Cohere said that Command A can deliver tokens at a rate of up to 156 tokens/sec, which is 1.75x faster than GPT-4o and 2.4x faster than DeepSeek-V3. With business use in mind, the model also has a larger context window at 256,000 tokens, which is twice the size of the industry average, including Cohere's Command R+ model. It means that the model can ingest a sizable number of documents at once or up to a 600-page book. "We are only training our model to make you better at your job," Cohere co-founder Nick Frosst said. "It should feel like getting into a mech for your mind. So, we are training it to empower you. So, it should feel specifically good at that." The company stated that it focused on developing capabilities in the model designed to enable the scalable operation of AI agents. Agentic AI has recently become a prominent trend in the industry, aiming to create artificial intelligence systems that can analyze data, make decisions, and carry out tasks with minimal or no human involvement. In practice, this requires massive amounts of compute and doing so efficiently and accurately based on company information requires well-trained AI models. Cohere said Command A would integrate directly with its secure AI agent platform, North, which allows enterprise business users to use the full potential of their company data. The platform is designed to enable enterprise AI agents to use customer relationship management, resource planning software and other tools to automate tasks.

Cohere's 111B-parameter AI model can run on just two GPUs

Cohere has released Command A, a high-performance AI model featuring 111 billion parameters, a 256K context length, and support for 23 languages, on March 16, 2025. The model is designed for enterprise applications, promising a 50% reduction in operational costs compared to existing API-based models. Command A addresses the significant challenges posed by training and deploying large-scale AI models that often require extensive computational resources. Typical models, such as GPT-4o and DeepSeek-V3, demand up to 32 GPUs and extensive infrastructure, which poses a barrier for smaller enterprises. Command A, however, operates efficiently on just two GPUs while maintaining competitive performance levels. The core architecture of Command A employs an optimized transformer design featuring three layers of sliding window attention, each with a window size of 4096 tokens. This structure enhances local context modeling, allowing the model to effectively manage detailed information across lengthy text inputs. Additionally, it includes a fourth layer that consists of global attention mechanisms, facilitating unrestricted token interactions throughout the entire sequence, thereby enriching its contextual understanding. Command A achieves a token generation rate of 156 tokens per second, which is 1.75 times faster than GPT-4o and 2.4 times faster than DeepSeek-V3. Its performance in handling instruction-following tasks, SQL queries, and retrieval-augmented generation (RAG) applications has shown exceptional accuracy in real-world evaluations, outperforming its competitors in multilingual scenarios. Baidu just made AI cheaper: Ernie 4.5 costs 1% of GPT-4.5 The model's multilingual capabilities extend beyond basic translation, exhibiting superior proficiency in various dialects, including Arabic, with evaluations showing enhanced contextually appropriate responses for regional dialects such as Egyptian, Saudi, Syrian, and Moroccan Arabic. This linguistic versatility is particularly beneficial for businesses operating in diverse language environments. Performance evaluations indicate that Command A consistently outperforms its peers in fluency, faithfulness, and response utility during human assessments. It is equipped with advanced RAG capabilities that include verifiable citations, which enhance its utility for enterprise information retrieval applications. Furthermore, the model includes high-level security features designed to protect sensitive business information. Noteworthy features of Command A include: The introduction of Command A marks a significant advancement for businesses seeking cost-effective, efficient AI solutions that maintain robust performance standards.

New Cohere AI model's performance rivals latest versions of DeepSeek, ChatGPT

TORONTO -- Cohere Inc. says it has released a new enterprise generative artificial intelligence model that is "on par or better than" GPT-4o and DeepSeek-V3. The Toronto-based tech firm says its new product Command A offers maximum performance with minimal hardware costs. Command A is well suited for agentic tasks, which tend to involve sophisticated reasoning or many complex steps. Cohere co-founder Nick Frosst says the technology is particularly good at helping users securely get responses to questions based on internal company data and tools. When combined with Cohere's North platform, he says it can also help users work more productively on tasks like analyzing long reports or summarizing data from sources like emails. The performance of Command A puts Cohere in the same league as OpenAI, which sparked a race to advance AI when it released its chatbot ChatGPT in November 2022, and DeepSeek, the Chinese chatbot whose debut earlier this year came with promises of better performance and lower costs than ChatGPT.