Cohere
Cohere is a leading AI platform for enterprise. It offers large language models (LLMs) and Retrieval-Augmented Generation (RAG) capabilities that are specifical...
Last verified:
What is Cohere?
Cohere is an enterprise AI platform that provides foundational large language models (LLMs) and AI solutions for developers and businesses to build secure, scalable applications. The platform offers the Command family of generative models (Command A+, Command A, Command R7B, Command R+, Command R) for text generation, conversational agents, summarization, and copywriting, along with specialized models like Embed for semantic text representations, Rerank for search result optimization, Cohere Transcribe for audio transcription, and the Aya family for multilingual and multimodal tasks.
Key features include retrieval-augmented generation (RAG) with citations, structured outputs in JSON format, tool use and function calling for agents, support for 49 languages, 128K context windows, deployment in virtual private clouds (VPC), on-premises options, and a Model Vault for maximum security. The platform emphasizes enterprise-grade security with multi-layered protection, SOC 2 certification, and data sovereignty, keeping customer data under their control without training on it.
Cohere is designed for enterprise teams in financial services, public sector, energy, technology, healthcare, manufacturing, and telecommunications who need secure AI that enhances human judgment without outsourcing data. Developers use it for building RAG pipelines, semantic search systems, text classification, clustering, conversational agents, tool-using agents, translation, and speech-driven workflows. The company also launched North in 2025, a turnkey AI platform for workplace productivity with agentic capabilities.
Cohere pricing
Pricing model: Freemium
Cohere offers a free Trial tier and pay-as-you-go Production pricing. Trial (Free): $0 with 100 API calls per minute, 1,000 calls per month, access to all models, no credit card required, limited to non-production use. Command R+: $2.50 per 1M input tokens and $10 per 1M output tokens. Command R: $0.15 per 1M input tokens and $0.60 per 1M output tokens. Command R7B: $0.15 per 1M input tokens and $0.0375 per 1M output tokens. Embed v3: $0.10 per 1M tokens. Rerank 3.5: $2.00 per 1,000 searches. Generative models priced per token with input and output priced differently. Rerank priced per search. Embed priced per token embedded. Enterprise deployments with custom SLAs, dedicated support, and fine-tuning require contacting sales for custom pricing.
Cohere pros
- Free trial with 1,000 API calls per month and 100 calls per minute
- No credit card required for trial access
- Access to all models including Command R+, Embed, and Rerank during trial
- Command R+ cheaper than Claude Sonnet and GPT-4.1 at $2.50/$10 per 1M tokens
- Command R extremely cost-effective at $0.15/$0.60 per 1M tokens
- Embed v3 at $0.10 per 1M tokens for high-quality multilingual embeddings
- Rerank 3.5 dramatically improves RAG accuracy at $2 per 1,000 searches
- 128K context window on Command R and Command R+ models
- Supports 49 languages for global applications
- Deploy in your own VPC or on-premises for maximum data control
- SOC 2 Type 2 certified with industry-standard security
- Structured outputs guarantee 100% JSON compliance with schema
- Advanced RAG with citations showing information sources
- Fine-tuning available for generative, chat, rerank, and classification models
- Compatible with AWS Bedrock, SageMaker, Azure, Oracle GenAI, and Google Cloud
Cohere cons
- No permanent free tier for production applications
- Free trial limited to non-production use only
- No pay-as-you-go middle ground between free and production pricing
- Command R+ output tokens cost 4x input tokens making long responses expensive
- Rerank charges per search not per document affecting batch costs
- Documents over 500 tokens get split into chunks increasing token usage
- Embedding large document collections adds up quickly in cost
- Command R+ may lag behind Sonnet 4.6 and GPT-4.1 on general reasoning and coding
- Enterprise pricing requires contacting sales with no transparent public pricing
- Trial limited to 1,000 calls per month not enough for production workloads
Frequently asked questions about Cohere
What is Cohere?
Cohere builds foundational models and AI solutions that help teams turn everyday effort into extraordinary impact. The company provides enterprise-grade large language models including the Command family for text generation, Embed for semantic text representation, Rerank for search optimization, and Cohere Transcribe for audio transcription. Cohere emphasizes security, allowing deployment in your own VPC or on-premises so your data stays under your control.
How much does Cohere cost?
Cohere has a free trial with 1,000 API calls per month and 100 calls per minute. For production, Command R+ costs $2.50/$10 per 1M input/output tokens, Command R costs $0.15/$0.60 per 1M tokens, Embed v3 costs $0.10 per 1M tokens, and Rerank v3 costs $2 per 1,000 searches. Input and output tokens are priced differently for generative models. Enterprise pricing requires contacting sales.
Does Cohere have a free tier?
Cohere has a trial tier that is free with 1,000 API calls per month and 100 calls per minute. It includes access to all models and requires no credit card. However, it is limited to non-production use for prototyping and evaluation. There is no permanent free tier for production applications, and you must move to pay-as-you-go pricing for production workloads.
What is Cohere Rerank?
Rerank is a model that takes a query and a list of documents and re-orders them by relevance. It is the fastest way to inject language model intelligence into an existing search system and dramatically improves RAG accuracy by filtering out irrelevant retrieved documents before they reach your LLM. It can rerank up to 100 documents per search and costs $2 per 1,000 searches.
What is the difference between Command R and Command R+?
Command R+ is Cohere's most capable text generation model with strong performance at RAG and tool use, a 128K context window, and multilingual support for 10+ languages. It costs $2.50/$10 per 1M input/output tokens. Command R is the budget text generation model good for classification and extraction, also with a 128K context window, costing $0.15/$0.60 per 1M tokens. Command R is comparable to GPT-4o-mini and very cost-effective at scale.
Can I deploy Cohere models in my own infrastructure?
Yes, Cohere offers multiple deployment options for security and compliance. You can deploy in your virtual private cloud (VPC), on-premises on your own infrastructure for fully private deployment, or use a dedicated Cohere-managed Model Vault. This is a key differentiator from competitors like OpenAI and Anthropic which are API-only. Contact sales for enterprise deployment pricing.
What is Retrieval-Augmented Generation (RAG) in Cohere?
RAG allows an LLM to access external data sources like the internet or company internal documentation, leading to better and more factual generations. Cohere's advanced RAG capabilities let you see the underlying query the model generates and include citations pointing to where information was found. The Chat endpoint works with or without RAG, and you can build powerful RAG systems using Cohere Embed and Rerank models together.
What languages does Cohere support?
Cohere's Command family supports 49 languages for global communication and discovery. The Aya Expanse model specifically covers 23 languages including Arabic, Chinese, Czech, Dutch, English, French, German, Greek, Hebrew, Hindi, Indonesian, Italian, Japanese, Korean, Persian, Polish, Portuguese, Romanian, Russian, Spanish, Turkish, Ukrainian, and Vietnamese. Cohere Transcribe supports 14 languages for audio transcription.
What is Structured Outputs in Cohere?
Structured Outputs is a feature that forces the LLM's response to strictly follow a schema specified by the user, generating structured data 100% of the time. This increases reliability in enterprise applications where downstream applications expect correctly formatted output. You can use JSON mode for valid JSON objects or JSON Schema mode for greater control over output structure, eliminating hallucinated fields and entries.
How do I get started with Cohere?
To get started, sign up for a Cohere account to get a trial API key with no credit card required. You get 1,000 API calls per month to test all models. Access the Playground to try models, read the API documentation at docs.cohere.com, and use the Python SDK or other language clients. For production usage, you'll move to pay-as-you-go pricing. For enterprise deployment options, contact sales.