When the author of "At...
When the author of "Attention Is All You Need" decided to build enterprise AI
You've probably heard of the Transformer architecture – it was proposed in a 2017 paper titled "Attention Is All You Need" and laid the foundation for virtually all large language models today. One of the eight authors of that paper, Aidan Gomez, co‑founded Cohere in 2019 alongside two University of Toronto alumni, Nick Frosst and Ivan Zhang.
Their goal was straightforward: to turn the potential of the Transformer architecture into AI tools that enterprises can actually trust and use.
How is it different from ChatGPT?
You may have used AI chat tools like ChatGPT or Claude, which are aimed at the general public. Cohere's positioning is completely different – it doesn't build consumer‑facing chatbots; instead, it focuses on delivering customised generative AI services for enterprise clients.
Simply put, Cohere works for businesses, not for individuals to chat with. Its core advantage lies in supporting private deployment with data security and customised model development. This means that sensitive corporate data doesn't need to be uploaded to the cloud – AI models can run on the company's own servers.
What are its core prod...
What are its core products?
Cohere's product portfolio covers the entire chain from models to application platforms:
1. Command series models: enterprise‑grade large language models
Command is Cohere's flagship family of generative models. The latest version, Command A+, released in May 2026, uses a Mixture of Experts (MoE) architecture with 218 billion total parameters, but only 25 billion are activated per inference – so it has very low compute requirements: it can run on just one B200 or two H100 GPUs.
It supports 48 languages, including all official EU languages, as well as Japanese, Arabic, Hindi, and more. Inputs support both text and images, with a 128K token context window and a maximum output of 64K tokens. Most notably, it is open‑sourced under the Apache 2.0 licence, allowing enterprises to freely download, modify, and deploy it.
2. North platform: AI productivity workstation
In 2025, Cohere offici...
In 2025, Cohere officially launched North, an AI productivity platform for enterprises. It provides a chatbot interface that lets employees query internal company documents and data in natural language; it can automatically generate visualised financial data charts and competitor research reports; and it can use AI agents to automate multi‑step business processes.
Under the hood, it runs a "variant" version of the Command model. According to Cohere, its response generation is 75% faster than GPT‑4o, and it performs better on multiple benchmarks including instruction‑following, tool usage, and SQL code generation.
3. Other capabilities
Cohere also offers Embed (vector embedding) and Rerank models for enterprise‑grade intelligent search and information retrieval, as well as Cohere Transcribe, an open‑source speech‑recognition model supporting 14 languages.
Why has it attracted so much capital backing?
Cohere's investor list is quite impressive: Nvidia, Salesforce Ventures, and Oracle are all strategic investors. According to public reports, Cohere has raised approximately $1.5 billion in total funding, with a valuation of around $20 billion.
Notably, in April 2026...
Notably, in April 2026, Cohere announced a merger with German AI company Aleph Alpha and is currently conducting a $600 million Series E funding round. Cohere has also been named to the Forbes AI 50 list for multiple consecutive years.
Who are its clients?
Cohere's clients come mainly from industries with extremely high data‑security requirements, such as finance and technology. Publicly disclosed customers include Fujitsu (Japan), LG CNS (South Korea), Dell Technologies, and the Royal Bank of Canada. In June 2026, Cohere also signed a $220 million AI cloud services contract with HIVEDigitalTech.
What is "Sovereign AI" that it's pushing for?
Cohere's current core strategic direction is "Sovereign AI" – enabling every country and every enterprise to own and control their own AI infrastructure, rather than renting APIs from a few US‑based companies.
The open‑sourcing of Command A+ is a manifestation of this strategy. It allows enterprises to deploy in VPCs, on‑premises servers, or even in completely air‑gapped environments, without transmitting any data to external systems. For governments, financial institutions, healthcare organisations, and other entities with strict data‑sovereignty requirements, this may be far more attractive than any cloud API.
To be honest
To be honest
Cohere's story is quite different from that of many AI companies. It didn't rush to build a free chatbot for everyone; instead, it has steadily constructed AI infrastructure that enterprise customers can deploy privately.
Its founder is one of the authors of the Transformer paper. It chose to open‑source its models. It coined the concept of "Sovereign AI." Behind all these moves lies a common logic: AI should not be just an API service provided by a handful of large companies – it should become a tool that every organisation can own and control.
If you're a business with stringent data‑security requirements, or if you're simply curious to see what distinguishes "enterprise AI" from "consumer AI," Cohere is worth taking some time to understand.