Cohere
Enterprise models and agents built for private deployment
Overview
Cohere is a Toronto-based enterprise AI company that builds the Command generation models, Embed and Rerank retrieval models, Parse and Transcribe, and the North agent platform. Its models run in Cohere's cloud, on the major hyperscalers, in dedicated Model Vault instances or fully on-premises. A combination with Germany's Aleph Alpha was signed in September 2026 and awaits regulatory approval.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- Free API calls
- Rate-limited
- Not for production or commercial use
- Command R7B $0.0375 in / $0.15 out per 1M tokens
- Command R $0.15 / $0.60 per 1M tokens
- Embed 5 Pro $0.12, Embed 5 Fast $0.08 per 1M tokens
- Rerank 4 Pro $2.50, Rerank 4 Fast $2 per 1K searches
- Parse 5 $1.50 per 1K pages
- Apache 2.0 license
- Runs on two H100 GPUs
- 48 languages
- Tool use and reasoning
- Fully managed dedicated deployment
- No shared resources or multi-tenancy
- Integration with North
- AI agents for routine and complex workflows
- Intelligent search
- Pre-built data connectors and document parsing
API pricing (pay as you go)
| Model | Input | Output |
|---|---|---|
| Command R | $0.15 | $0.60 |
| Command R7B | $0.0375 | $0.15 |
| Command (legacy) Legacy model, pricing for existing customers (FAQ) | $1 | $2 |
| Command-light (legacy) Legacy model, pricing for existing customers (FAQ) | $0.30 | $0.60 |
| Command R 03-2024 (legacy) Legacy model, pricing for existing customers (FAQ) | $0.50 | $1.50 |
| Command R+ 08-2024 (legacy) Legacy model, pricing for existing customers (FAQ) | $2.50 | $10 |
| Command R+ 04-2024 (legacy) Legacy model, pricing for existing customers (FAQ) | $3 | $15 |
| Aya Expanse (8B / 32B) | $0.50 | $1.50 |
Use cases
What you can produce with Cohere
- Search and RAG pipelines built on Embed 5 vectors and Rerank 4 re-scoring
- A self-hosted Command A+ deployment under Apache 2.0, on as little as two H100 GPUs
- Structured data extracted from PDFs, forms, tables and handwriting with Parse 5
- Workplace agents and automations built in North 2, including air-gapped installations
- Dedicated, single-tenant model inference through Model Vault
- Multilingual generation and translation across 48 languages
Cohere suits enterprises that need to keep AI inside their own infrastructure. Embed 5 and Rerank 4 are practical building blocks for search and retrieval-augmented generation, Command A+ is now open-weight under Apache 2.0, and North 2 adds agents with cost controls and air-gapped deployment.
There is no consumer chatbot, the flagship Command A has no public per-token price, and the Aleph Alpha combination has not yet closed. A good shortlist candidate for regulated ASEAN organisations that need on-premises deployment; individuals have little reason to use it.
What we found
Cohere is a Toronto company, started in 2019, that sells language models and an agent platform to enterprises rather than to consumers. Its retrieval models are the core of its reputation: Embed 5, launched on 30 September 2026 in Pro and Fast versions with a 128K context and text, image and mixed inputs across more than 100 languages, and the Rerank 4 models. Generation runs on the Command family. Command A+, released on 20 May 2026 under Apache 2.0, is a mixture-of-experts model with 218 billion total and 25 billion active parameters, 48 languages, image input and a 128K context, and Cohere says it runs on two H100 GPUs. North Mini Code, an open agentic coding model, followed in June.
North, the workplace agent platform Cohere launched in 2025, reached North 2 on 5 October 2026, adding a new agent harness, support for other vendors' models, cost controls and air-gapped deployment. Deployment choice is the selling point: Cohere's cloud, the major hyperscalers, dedicated Model Vault instances, private VPCs or on-premises.
The corporate picture is still settling. Cohere and Germany's Aleph Alpha announced a planned partnership in April and signed a definitive business combination on 16 September 2026; the deal still needs regulatory approval, and the combined company plans dual headquarters in Toronto and Berlin. Cohere's last round it announced itself was $500 million at a $6.8 billion valuation in August 2025; the roughly $20 billion figure for the combined company comes from press reports, not from Cohere.
Pricing is partly public. A trial API key is free but rate-limited and barred from production or commercial use. Production keys bill per use: Command R7B at $0.0375 in and $0.15 out per million tokens, Embed 5 Pro at $0.12 per million, Rerank 4 Pro at $2.50 per thousand searches. Command A carries no list price on the page, and North is quoted. For ASEAN firms with residency rules, the on-premises options matter more than the token rates.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Notable facts
- Cohere co-founder Aidan Gomez was a co-author of the 'Attention Is All You Need' paper that introduced the Transformer architecture, which underpins every modern LLM.
- Cohere's models can be deployed entirely within a customer's own AWS or Azure environment, with data never leaving the enterprise boundary.
- The company's Command R+ achieved first place on the Berkeley Function-Calling Leaderboard, outperforming GPT-4 on structured tool use tasks.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Cohere's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Cohere unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Cohere directly →
Spotted something out of date? Suggest an update →
Alternatives to Cohere
More in LLMs & Chat
