Overview
Cohere is an enterprise-focused large-language-model provider that positions itself around retrieval, multilingual capability, and deployment flexibility rather than consumer chat. Its product line spans the Command family of generation models, Embed for multilingual embeddings, and Rerank for improving retrieval relevance, alongside the North agent platform and Compass enterprise search. A distinguishing fact for buyers in regulated industries: Cohere supports private cloud, virtual private cloud, and on-premise deployment, and states that it does not train on customer data.
The company was founded in 2019 and is headquartered in Toronto, with roughly 450 employees and a co-founder, Aidan Gomez, who was among the authors of the original transformer research. A $100 million round extension in September 2025 lifted its valuation to about $7 billion, and the company reported reaching roughly $100 million in annualised revenue during 2025. Cohere models are also available through Amazon Bedrock, Google Vertex AI, Microsoft Azure, and Oracle Cloud Infrastructure, which lets enterprises adopt them without a direct billing relationship. It competes with OpenAI, Anthropic, Google, and Mistral, and differentiates on enterprise deployment and retrieval rather than frontier-benchmark leadership.
Key Features
- Command R and Command R+ generation models for retrieval-augmented use cases
- Command R7B smaller model for low-latency and lower-cost workloads
- Embed multilingual embedding models for search and classification
- Rerank models that re-order retrieval results for relevance
- North agent platform for building task automation over enterprise data
- Compass enterprise search across structured and unstructured sources
- Private cloud, VPC, and on-premise deployment for regulated data
- Availability on Amazon Bedrock, Google Vertex AI, Azure, and OCI
- Fine-tuning of generation and embedding models
- Tool use and function calling for agentic workflows
- Strong multilingual coverage across many languages
- Stated policy of not training on customer data
Pricing
| Tier | Model | Typical Cost | Included |
|---|---|---|---|
| Command R | Per 1M tokens | $0.15 input / $0.60 output | General generation and RAG |
| Command R+ | Per 1M tokens | $2.50 input / $10.00 output | Complex reasoning and synthesis |
| Rerank 3.5 | Per search unit | ~$0.001–$0.0025 | Retrieval result reranking |
| Embed | Per 1M tokens | ~$0.10–$0.12 | Multilingual embeddings |
| North / Compass / Dedicated | Quote | Contact for quote | Agent platform, search, private deployment |
Pricing verified June 2026. Enterprise pricing requires a quote.
Strengths
- Deployment flexibility including VPC and on-premise for regulated industries
- Strong retrieval stack pairing Embed and Rerank with the Command models
- Broad multilingual coverage suited to global enterprises
- Cloud-agnostic availability across the major hyperscaler model marketplaces
- Clear stance that customer data is not used for training
Limitations
- Frontier reasoning and coding benchmarks trail OpenAI, Anthropic, and Google models
- Smaller developer community and third-party ecosystem than the largest providers
- North and Compass have no published pricing and require a sales conversation
- Fewer multimodal capabilities than rivals with image and audio models
- Lower brand recognition than hyperscaler-backed competitors can slow internal buy-in
Buyer Considerations
Aggregate sentiment frames Cohere as a pragmatic enterprise choice rather than a consumer or hobbyist favourite. Buyers in financial services, healthcare, and the public sector value the private and on-premise deployment options and the explicit data-handling commitments, and reviewers consistently rate the Embed and Rerank components highly for production retrieval quality. The availability of models through Bedrock and Vertex AI is cited as a way to adopt Cohere within existing cloud governance.
The reservations are predictable for a focused challenger. Reviewers note that for open-ended reasoning or coding they still reach for frontier models from OpenAI, Anthropic, or Google, and that Cohere's community, sample code, and integrations are thinner. The lack of published pricing for North and Compass draws criticism from teams that want to evaluate before engaging sales. Overall sentiment is positive on retrieval and deployment, mixed on raw model breadth. Paraphrased from aggregate review themes.