Cohere

Cohere is a Canadian artificial intelligence company that develops large language models (LLMs) and AI infrastructure products exclusively for enterprise customers. Unlike consumer-facing AI companies, Cohere has no public chatbot product; its entire business is built around providing LLMs and deployment tooling to organisations in regulated sectors including finance, healthcare, manufacturing, and the public sector. This enterprise-only focus, combined with strong data security and private deployment options, has positioned Cohere as a preferred AI provider for organisations with strict data governance requirements.

Founding and History

Cohere was founded in 2019 by Aidan Gomez (CEO), Ivan Zhang, and Nick Frosst. Gomez was a co-author of the seminal "Attention Is All You Need" paper (2017) that introduced the transformer architecture — the foundation of all modern large language models — while at Google Brain. This research pedigree gave the company early credibility in enterprise AI circles. Headquartered in Toronto, Cohere also has offices in San Francisco, London, and other cities, and has grown to several hundred employees.

Products and Model Families

Command

The Command model family is Cohere's core generative AI offering, handling tasks such as text generation, summarisation, classification, question answering, and code generation. Command models are designed for enterprise reliability — predictable outputs, controllable tone, and low hallucination rates on grounded tasks. Command R and Command R+ (2024) were optimised for retrieval-augmented generation (RAG) tasks, with strong performance at citing sources and reasoning over retrieved documents. In August 2025, Cohere released Command A Translate, a specialised 111-billion-parameter translation model achieving state-of-the-art performance across 23 languages, targeting enterprises with multilingual content workflows.^[1]

Embed

Cohere's Embed models convert text into dense vector representations for semantic search, document retrieval, and clustering. Embed v4, released in 2025, is a multimodal embedding model supporting both text and image inputs with Matryoshka Embedding support — allowing vectors to be truncated to smaller dimensions without significant quality loss — and coverage of over 100 languages. Embed models are a cornerstone of enterprise RAG deployments where accurate retrieval from large document repositories is critical.

Rerank

The Rerank model takes a query and a set of candidate documents (retrieved by any method, including keyword or semantic search) and rescores them for relevance to the query. This two-stage retrieval pattern — retrieve with embedding similarity, refine with reranking — consistently outperforms single-stage retrieval in enterprise search applications.^[2]

North

Launched in January 2025, North is Cohere's turnkey enterprise AI platform for workplace productivity.^[3] North provides an agentic interface that allows employees to query connected data sources, automate workflows, and surface insights from enterprise data — all while keeping data within the organisation's secure environment. North competes with Microsoft 365 Copilot and Salesforce Einstein Copilot in the enterprise AI assistant category.

Model Vault

Introduced in September 2025, Model Vault is Cohere's dedicated inference infrastructure for enterprises requiring the highest levels of data sovereignty. It deploys Command, Rerank, and Embed models within isolated Virtual Private Clouds (VPCs) or on-premises environments, ensuring that inference requests and data never traverse shared infrastructure. This product targets financial institutions, defence contractors, and healthcare organisations subject to stringent data residency regulations.

Business Performance

Cohere's revenue trajectory has been steep: from approximately USD 35 million in annualised recurring revenue (ARR) at the start of 2025 to USD 240 million by year-end, representing over 50% quarter-on-quarter growth through the year.^[4] CEO Aidan Gomez publicly stated in October 2025 that an IPO was imminent, and the appointment of IPO-experienced CFO François Chadwick signalled preparations for a public listing, widely anticipated for 2026. The company's cloud-agnostic strategy — offering models on AWS, Azure, GCP, and through OCI as well as private deployments — has been central to its enterprise sales success, allowing customers to deploy Cohere's models wherever their data already resides.

Competitive Position

Cohere competes with OpenAI (Azure OpenAI Service), Anthropic (Claude for Enterprise), Google (Vertex AI), and AI21 Labs in the enterprise LLM market. Its differentiation centres on data privacy, sovereign deployment, multilingual capability, and its model-agnostic deployment story. Cohere does not require customers to send data to a public API — a meaningful advantage in regulated industries. Cohere's enterprise positioning aligns closely with the requirements of Malaysian regulated sectors, particularly banking and financial services, where data sovereignty, PDPA compliance, and BNM's Risk Management in Technology (RMiT) framework impose strict constraints on how AI providers may handle customer data. The ability to deploy Cohere models within a private VPC or on-premises via Model Vault is directly relevant to the compliance architectures that Malaysian banks including Maybank, CIMB, Public Bank, and RHB must maintain. Malaysia's financial regulator, Bank Negara Malaysia (BNM), has issued guidance under its Financial Sector Blueprint 2022–2026 that requires financial institutions to assess AI risk, maintain explainability, and ensure data does not leave approved jurisdictions without appropriate controls. Cohere's sovereign deployment infrastructure addresses these requirements in a way that public-API-only LLM providers cannot. Similarly, the Securities Commission Malaysia (SC) has articulated expectations around the use of AI in capital markets activities that favour verifiable data controls. In the healthcare sector, Kementerian Kesihatan Malaysia (KKM, the Ministry of Health) is evaluating AI tools for clinical decision support and medical record summarisation. The sensitivity of patient data under Malaysia's PDPA means that healthcare providers strongly prefer AI vendors offering private deployment — a factor that favours Cohere's Model Vault approach over consumer-grade API services. Cohere's multilingual Embed v4 and Command A Translate models have potential relevance to Malaysian organisations managing content in Bahasa Malaysia, English, and Chinese. Malaysian government agencies with multilingual service mandates — including MAMPU, DBKL, and statutory bodies — could benefit from enterprise translation and semantic search across language boundaries. Malaysian AI distributors and cloud resellers in the Microsoft Azure and AWS partner ecosystem are positioned to bring Cohere's products to local enterprise customers. As Malaysia's AI ecosystem matures under the National AI Roadmap, enterprise LLM platforms that meet data governance requirements are increasingly preferred over free-tier AI tools for production business applications. MDEC's Malaysia Digital Acceleration Grant (MDAG) programme has flagged enterprise AI infrastructure as an eligible investment category, which may support local adoption of platforms like Cohere.

1. Cohere. (2025). Command A Translate: State-of-the-art enterprise translation. Cohere Blog.
2. Cohere. (2024). Rerank 3: A new foundation for enterprise search. Cohere Documentation.
3. SiliconANGLE. (2025). Cohere introduces LLM-powered North productivity platform. SiliconANGLE Media.
4. n1n.ai. (2026). Cohere path to IPO following $240 million revenue milestone. n1n.ai Research.