Cohere

Cohere develops large language models and supporting infrastructure that let developers and enterprises build AI applications. Its model portfolio includes the Command family of generative models (Command A+, Command A, Command R+, Command R and Command R7B), Embed for semantic representation and retrieval, Rerank for search ranking, Transcribe for speech recognition, Parse for document parsing, Aya multilingual models, and North Mini Code for agentic coding. Command models are described as supporting 49 languages with unified reasoning, tool orchestration and multimodal capabilities; Transcribe supports 14 languages.

The company's application layer centers on North, an agentic AI platform for enterprise work, and Compass for search and discovery. Models are reachable through the Cohere API and the Chat, Embed, Classify and Rerank endpoints, support retrieval-augmented generation with citations, and can be fine-tuned on customer data. Deployment options span the Cohere-hosted API, Model Vault (dedicated single-tenant inference, with an Encrypted variant using confidential computing and remote attestation), private deployments in a customer VPC or on-premises including air-gapped environments, and cloud AI platforms such as Oracle GenAI Service, AWS Bedrock and SageMaker, Google Cloud and Azure AML.

Cohere also operates a research lab, Cohere Labs (previously referred to as Cohere For AI), which publishes papers, releases open multilingual models such as Aya, and runs initiatives including Research Scholars, an open science community and Catalyst Grants. Developer resources include documentation, a Playground, LLM University and a Discord community.

Founding story

Aidan Gomez launched Cohere in 2019 with Nick Frosst and Ivan Zhang, following research the group had done together at FOR.ai, a precursor to the company. Gomez is a co-author of the 2017 paper "Attention Is All You Need" that underpins modern generative AI models.

Business model

Cohere sells access to its foundation models and enterprise applications to businesses and governments, through a hosted API with published pricing, dedicated managed inference (Model Vault), private and on-premises deployments, and distribution via major cloud AI marketplaces. Customer engagements include customization and fine-tuning of models on proprietary data.

Revenue comes from enterprise sale of model access and AI applications, including API usage, dedicated and private deployments, and customized models; the company reported roughly $35M in annualized revenue at the end of March 2024, up from about $13M at the end of 2023.

Traction

Annualized revenue reached about $35M at the end of March 2024, up from roughly $13M at the end of 2023, with a customer base of hundreds of companies. Total capital raised reached $970M as of July 2024, followed by a further $500M in August 2025 at a $6.8B valuation. Headcount was about 250 in mid-2024, with plans reported to double that year. Strategic partnerships include Oracle, Dell, RBC, Bell, Fujitsu, LG CNS, SAP and Ensemble Health Partners, and CoreWeave has deployed Cohere North for customer support.

Latest developments

Recent activity includes the launch of North Automations for workflow orchestration (July 2026), the release of Cohere Transcribe Arabic (July 2026), a partnership with the University of Toronto on responsible AI adoption (July 2026), a University of Waterloo partnership on Canada's AI talent pipeline, publication of research on the state of sovereign AI adoption in 2026, and the introduction of Model Vault for isolated dedicated inference.