Coding & Development

Cohere

Secure, enterprise-grade LLMs built for the real business world

Free rate-limited trial API key (non-commercial). Production API is pay-as-you-go; the pricing page lists rates only for legacy models (e.g. Command R+ 08-2024 $2.50 in / $10 out per 1M tokens). Model Vault dedicated GPUs are priced per hour, from $1/hr (L4, promo to Mar 2027) to $10/hr (GB200 promo). North, Compass and private deployments are custom.
Visit Cohere →
Pricing
Free rate-limited trial API key (non-commercial). Production API is pay-as-you-go; the pricing page lists rates only for legacy models (e.g. Command R+ 08-2024 $2.50 in / $10 out per 1M tokens). Model Vault dedicated GPUs are priced per hour, from $1/hr (L4, promo to Mar 2027) to $10/hr (GB200 promo). North, Compass and private deployments are custom.
Best for
Who it is for: enterprise and public-sector teams with data-residency or security requirements, and developers building retrieval pipelines who want embeddings and reranking from one vendor.
Official site
cohere.com
Last checked
2026-11-11

Cohere is a strong choice for enterprises that need private or dedicated deployment of language, embedding and reranking models, especially for search and retrieval-based apps. Skip it if you want transparent self-serve pricing, a big hobbyist community, or the cheapest pay-per-token chatbot API. The company, founded in Toronto in 2019, sells Command models for generation, Embed and Rerank for retrieval, and products such as North (an agent platform) and Model Vault (isolated, dedicated inference). Models can be used through a production API, dedicated instances, or on-prem and VPC deployments. I could not verify current Command per-token prices, since the pricing page lists only legacy models; contact Cohere for current rates.

Best for

Who it is for: enterprise and public-sector teams with data-residency or security requirements, and developers building retrieval pipelines who want embeddings and reranking from one vendor. Who it is not for: solo developers or small teams wanting fully public pricing, anyone needing commercial use on a free key, or those seeking the largest community and tooling ecosystem.

Key features

01

Command models

Generative models for chat, tool use and retrieval-augmented generation. The homepage does not name specific versions, so check the docs.

02

Embed

Embedding models for semantic search; the homepage announces Embed 5.

03

Rerank

Reorders search results for relevance; priced per search unit (one query with up to 100 documents).

04

North

An agentic enterprise AI platform customers can own and deploy; custom pricing.

05

Model Vault

Dedicated, isolated inference with end-to-end encryption, billed hourly or via monthly/annual commitments.

06

Private deployments

On-premises or isolated VPC deployment, with customization at custom pricing.

07

Other models

Transcribe (speech), Translate and Parse (documents) are listed on the homepage.

Pricing breakdown

Trial API key

$0
Free, rate limited
  • Not for production or commercial use

Production API

Pay-as-you-go
Per 1M tokens; billed at month end or when balance reaches $250
  • Command R+ 08-2024: $2.50 input / $10 output
  • Aya Expanse 8B/32B: $0.50 input / $1.50 output
  • Current Command, Embed and Rerank prices not listed on the page

Model Vault

From $1.00/hr per GPU (promo)
Hourly per GPU; monthly/annual commit via sales
  • L4 $1.00/hr promo (standard $3.00)
  • A100 $2.25/hr promo
  • H100 80GB $4.25/hr promo (standard $6.25)
  • H200 $5.50, B200 $8.00, GB200 $10.00 per hour promo
  • Promo prices run to Mar 31, 2027

North, Compass, private deployment

Custom
Contact sales
  • Agent platform and enterprise search
  • On-prem or VPC deployment

Pros and cons

Pros

  • Deployment range from API to dedicated Model Vault to on-prem or VPC suits regulated organizations
  • Embed and Rerank give a complete retrieval stack from one vendor
  • Free trial key lets developers test before committing
  • North packages agents over company data, reducing integration work
  • Model Vault GPU rates are published per hour, giving some cost transparency for dedicated hosting

Cons

  • The pricing page shows only legacy Command per-token prices, so current production costs cannot be checked without contacting Cohere
  • North, Compass and private deployments need a sales demo
  • Dedicated hosting is billed per GPU-hour (from $1/hr for an L4 on promo to $10/hr for a GB200), so an always-on instance adds up fast; promo rates end March 31, 2027.
  • Trial keys are rate limited, non-commercial, and production access requires an application
  • Smaller developer community and third-party ecosystem than the largest model providers

Alternatives to Cohere

Frequently asked questions

Is there a free way to try Cohere?

Yes, a free trial API key, which is rate limited and cannot be used for production or commercial work.

How is the production API billed?

Pay-as-you-go, billed at month end or when the balance reaches $250. Current per-model rates were not clearly listed on the pricing page, so confirm in your dashboard or with sales.

What does a dedicated Model Vault cost?

It is now priced per GPU per hour. Promo rates (to March 31, 2027) run from $1.00/hr for an L4 to $10.00/hr for a GB200; monthly or annual commitments go through sales.

Can I deploy on my own infrastructure?

Yes, Cohere offers private deployments on-premises or in isolated VPCs, at custom pricing.

Is it a good fit for RAG?

Embed and Rerank are built for retrieval pipelines and are its main strength; test them against alternatives on your own data.

Ready to try Cohere?

Head to the official site to explore pricing and start a free trial where available.

Visit Cohere →