Cohere

Most foundation-model companies want your data flowing through their cloud. Cohere built the opposite pitch: run our models inside your own walls, on your VPC or on-prem, and we never see what goes in or comes out. That bet on data sovereignty is the whole company, and it is why banks, hospitals, and governments keep showing up on the customer list.

Founded in 2019 in Toronto by Aidan Gomez, Nick Frosst, and Ivan Zhang, Cohere has raised close to $1.5 billion and now carries a $7 billion valuation. It crossed $240 million in annual recurring revenue in 2025 and is openly positioning for an IPO, which is a rare thing to say out loud in a market crowded with OpenAI and Anthropic.

Cohere sells two things that fit together. One is a family of frontier models: Command for generation and agentic reasoning, Embed and Rerank for retrieval, and Transcribe for speech. The other is North, a turnkey AI workspace that stitches those models into agents, search, and document tools your team actually uses at work.

The common thread is control. Every product can run privately, be trained on your proprietary data, and stay auditable, which is the part enterprises in regulated industries care about more than raw benchmark scores.

The hard part of enterprise AI is not the demo, it is deploying something useful without shipping sensitive data to a third party. Cohere's answer is private deployment inside a customer's own cloud, on-prem environment, or a Cohere-managed Model Vault, so the data never leaves. North layers on secure, grounded retrieval across code, docs, tickets, and CRM records, then lets agents actually act: triaging issues, drafting release notes, running due diligence, and automating back-office work.

The payoff Cohere points to is accuracy in places where being wrong is expensive. One banking customer put it bluntly: they do not need a model that talks like Shakespeare, they need financial advice that is correct every time.

Cohere aims squarely at large, regulated organizations: financial services, healthcare, manufacturing, telecom, energy, technology, and the public sector. These are buyers who will trade a little frontier flash for auditable outputs, private deployment, and a compliance story.

The pitch lands with two audiences at once. Business and ops leaders get North's workplace automation, while developers get direct API access to Command, Embed, and Rerank for building their own RAG and agentic pipelines.

Cohere's lineup splits cleanly into models and platforms. Command handles generation and agentic work, Embed and Rerank do the heavy lifting for search and RAG, and Transcribe covers speech. Rerank is the quiet standout: it sits between retrieval and the model, scoring documents for true relevance, and it is the piece reviewers most often single out.

On top sit North, the AI workspace for agents and automation, and Compass, an enterprise search system that indexes noisy, multilingual, multimodal data so teams can actually find answers across it.

Cohere makes money two ways. North and Compass are sold on custom enterprise pricing, negotiated per deployment, which is where the big regulated accounts live. The models are more self-serve: a free, rate-limited trial API key to start, then pay-as-you-go token billing once you move to a production key.

For customers who want dedicated capacity, Model Vault runs Cohere models on managed instances priced per instance and performance tier, billed hourly or on longer commitments. It is a classic land-and-expand shape: start on the API, graduate to enterprise contracts and private deployment.

Cohere competes with OpenAI, Anthropic, and Mistral, and it does not pretend to win the raw-reasoning leaderboard. Reviewers note that GPT and Claude edge out Command on general benchmarks. Cohere's counter is deployment: VPC and on-prem options the big labs largely do not offer, a genuinely differentiated Rerank model for RAG, and cost-efficient retrieval at scale.

The strategic bet now is the agentic enterprise, anchored by North and reinforced by hardware and cloud partnerships including AMD, NVIDIA, Oracle, and the major clouds.

The traction is real and named. Cohere's technology is embedded across more than 100 generative AI use cases in Oracle Fusion Cloud Applications, and it has piloted or deployed North with RBC, Dell, LG, Palantir, Ensemble Health Partners, and Bell.

The business numbers back the logos: revenue crossed $100 million ARR in May 2025 and hit roughly $240 million by year end, with quarter-over-quarter growth above 50% through 2025. Custom models add technical proof too, like an 83.9% accuracy jump on HumanEval for code-tuned models.

The recurring praise is consistent: Cohere wins on deployment flexibility that the big labs will not match, Rerank is treated as best-in-class for RAG, and Embed is priced to compete for retrieval at scale. Enterprise buyers also value the security posture, with SOC 2 Type II, ISO 27001, and GDPR coverage.

The recurring knock is just as clear. For pure reasoning on general tasks, GPT and Claude are rated ahead of Command, so Cohere tends to win on fit and control rather than sheer model horsepower.

Cohere raised an oversubscribed $500 million round in August 2025 at a $6.8 billion valuation, led by Radical Ventures and Inovia Capital, with AMD Ventures, NVIDIA, PSP Investments, Salesforce Ventures, and HOOPP joining. A month later a $100 million extension nudged the valuation to $7 billion alongside an AMD partnership.

Added to its earlier Series A through D rounds, that brings total funding to nearly $1.5 billion. With ARR climbing past $240 million, Cohere is one of the few in this cohort talking openly about going public.

Cohere has picked a lane and committed to it: not the flashiest model, but the one a bank or a government is actually allowed to run. With $240 million in ARR, a $7 billion valuation, and North as the centerpiece, the growth is real and the IPO talk is not idle.

The risk is the same as the strategy. If OpenAI and Anthropic close the deployment and privacy gap, Cohere's edge narrows to Rerank and price. For now, sovereignty plus a strong retrieval stack is a defensible place to stand.

Cohere started as a Toronto research shop and still carries that DNA: it co-headquarters out of Toronto and San Francisco, runs offices in New York, London, Montreal, Paris, and Seoul, and operates remote-friendly across time zones. Its open-science arm, Cohere Labs, has grown to 4,500+ community members and 100+ published papers.

The culture leans research-first and high-agency, valuing autonomy, openness, and momentum. Engineering JDs ask for people who act, tinker, and drive change without waiting for perfect instructions, and who treat correctness and reliability as their own responsibility.

Cohere discloses pay across a genuinely global footprint, so its bands span USD, CAD, EUR, and more, reflecting roles from Toronto to London to the Middle East. US engineering roles run to the high six figures, while sales, marketing, and operations bands sit lower and vary by region.

Equity is part of the package across roles, and the company frames employees as co-owners. Benefits are unusually generous, including six weeks of paid vacation and six months of fully paid parental leave.

The registered entity is Cohere Inc., and its whole security story is that private and third-party deployments stay walled off. Cohere states it has no access to data processed in private deployments or on third-party managed AI services like AWS, Azure, or GCP, and it does not train on enterprise customer data.

The certifications back it: SOC 2 Type II, ISO 27001, ISO 42001 for responsible AI, plus GDPR coverage with Standard Contractual Clauses for EU data. A public Trust Center lets customers pull the actual reports.

The last year of coverage tracks Cohere's move from lab to IPO candidate. North launched publicly in August 2025 as its flagship agent platform, the funding rounds pushed the valuation to $7 billion, and by early 2026 the story turned to a $240 million revenue year setting the stage for a public listing.