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.

References

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.