Cohere

Cohere is an enterprise-focused AI platform providing large language models, embedding models, and search re-ranking tools built specifically for organizations that need data privacy, private deployment, and regulatory compliance. Founded in Toronto in 2019 by Aidan Gomez, Nick Frosst, and Ivan Zhang, three researchers with roots in Geoffrey Hinton's Google Brain lab, Cohere was built on unusually strong technical pedigree: Gomez was a co-author of the original 2017 "Attention Is All You Need" paper that introduced the Transformer architecture underpinning virtually all modern large language models. The company has raised over $934 million in venture capital, reached roughly $100 million in annual recurring revenue by mid-2026, and deliberately avoided acquisition offers, with Gomez stating publicly that "acquisition is failure."

Cohere's product suite centers on three model families: Command, its generative model line built for agentic, multilingual, and tool-use tasks across 49 languages, with the flagship Command A+ released open-weight in May 2026 and marketed for "sovereign critical infrastructure"; Embed, for multimodal semantic search and document retrieval; and Rerank, which boosts search relevance by re-scoring retrieved results against user intent. These three models are designed to work together as a complete retrieval-augmented generation (RAG) pipeline, a workflow Cohere has built its enterprise reputation around since its earliest days. In 2026, Cohere also launched North, a secure enterprise AI workplace platform that connects to a company's internal tools and data to power agent-based automation, positioning Cohere beyond pure API access into a full workplace productivity layer.

What sets Cohere apart from consumer-facing competitors like OpenAI or Anthropic is its singular enterprise focus: no consumer chatbot, no free-for-all app, just infrastructure for regulated industries including financial services, healthcare, energy, manufacturing, telecommunications, and the public sector. Deployment flexibility is central to this pitch, with options spanning public cloud API access, a customer's own VPC on AWS, Azure, or Oracle, fully on-premises hardware, or Cohere's dedicated, Cohere-managed Model Vault. Customers including Oracle, Dell, RBC, Fujitsu, SAP, Salesforce, TD Bank, McKinsey, and Accenture use Cohere's models for use cases where data cannot leave a specific security perimeter.

Pricing is genuinely two-tiered: a transparent, pay-as-you-go public API for the Command R and Command R+ model generation, priced competitively against equivalent OpenAI and Anthropic offerings, alongside custom enterprise agreements for the newest Command A+ family and any private or on-premises deployment, which require contacting sales directly. This makes Cohere best suited for enterprises and regulated organizations that specifically need data sovereignty, private infrastructure, and a RAG-first approach to enterprise search and knowledge retrieval, rather than developers or startups simply looking for the highest-benchmarking general-purpose model at the lowest published price.

Cohere runs a two-track pricing model. Its public, pay-as-you-go API charges per million tokens: Command R+ costs $2.50 (input) / $10.00 (output), Command R is $0.15/$0.60, and the economical Command R7B is $0.0375/$0.15. Embed v3 is priced at $0.10 per million input tokens, and Rerank v3 costs $2.00 per million tokens of search input processed. Command A, the newer general-purpose flagship, is priced at $2.50 input / $10.00 output per million tokens. Newer top-tier models, including Command A+, Command A Reasoning, Command A Translate, and Command A Vision, do not have public per-token pricing and require contacting Cohere sales; trial API keys for these are capped at 20 requests/minute and 1,000 calls/month. Enterprise and private deployment pricing (VPC, on-premises, or Cohere-managed Model Vault) is fully custom. On AWS Bedrock, Command Provisioned Throughput costs approximately $49.50/hour per model unit, or roughly $29,000/month, a meaningfully higher cost tier than the standard pay-as-you-go API.