Skip to content
AgentsIndex

Cohere

Enterprise model family with private and on-prem deployment, sold alongside its own agent platform.

Best for
Enterprise teams that need secure AI for search, agents, and private deployments.

What we checked

Pages last read
Facts checked
5 facts

From 7 sources, read 1 November 2026

AI answers
Too few read

58 buyer-question answers read, November 2026

At a glance

Tags
Private AI

For developers

Works with
Slack, Salesforce, Notion, Discord, Google Drive, GitHub and 16 more

Cohere in our research

All research

Study

AI Tool Pricing Index, November 2026

Pricing

Usage-based

43 of the 361 tools we checked bill by usage.

Edition price, checked 12 November 2026.

Pricing pages of the AI tools, November 2026 editionas of n = 361 tools checked

What is Cohere?

Cohere is an enterprise AI platform for teams that need secure search, agents, and private deployments across workplace systems. It combines North, Compass, Command, Transcribe, Embed, and Rerank to support generative workflows, document retrieval, speech transcription, and search quality. Customers include Oracle, Dell Technologies, RBC, and Asana. Pricing includes North and Compass on custom plans, plus Model Vault options starting at $2,000/month.

What Cohere is today

Cohere, founded in Toronto in 2019 by Aidan Gomez, Nick Frosst and Ivan Zhang, builds two platform products and a model lineup behind them. North is its agentic workplace-AI product (search, generation, automations, agents); Compass is a search-and-discovery layer with pre-built connectors, document parsing and a managed index. The models underneath are Command (generation), Embed (embeddings, now on version 5), Rerank (retrieval ranking), Transcribe (speech-to-text), and two that didn't exist when this listing was last checked: Parse (document parsing, launched as a standalone model) and North Small Translate (machine translation). Aya, Cohere Labs' multilingual research model, is still reachable through the API at its own per-token rate but now sits under Cohere's research section rather than its product nav.

Pricing: two products are sales-only, one has a real price list

North and Compass both show "Get in touch for custom enterprise pricing" on cohere.com/pricing -- there is no public number for either of Cohere's two flagship platform products. Model Vault, its dedicated-inference product, is the one place Cohere publishes real figures: hourly and monthly per-instance rates across Embed 5 (Fast/Pro, Small/Medium), Rerank (3.5, 4 Fast, 4 Pro), and Parse 5, running from $2,000/month (Embed 5 Fast or Pro, Small tier) to $6,500/month (Rerank 4 Pro, Large tier), billed hourly, monthly or on longer commitments. Direct pay-as-you-go per-token pricing for current-generation models isn't on that page at all -- the only per-token figures listed are for superseded models (Command, Command-light, Command R, Command R+ 04-2024), explicitly marked "for existing customers." And for several current Command A variants (Command A+, Command A Reasoning, Command A Translate, Command A Vision), Cohere's own docs say production API access requires emailing sales even though a trial key exists -- it's a trial-only tier, not self-serve production, for exactly the newest models a buyer is most likely to want.

Security and private deployment

Cohere states its API platform is SOC 2 Type II compliant, with annual third-party audits and penetration tests, plus a public bug-bounty program. For teams that need the data to never leave their own infrastructure, it offers three private-deployment paths: your own VPC, on-premises, or Model Vault (Cohere-managed but isolated, no shared resources or multi-tenancy). A public status page exists at status.cohere.com, and a trust center at trustcenter.cohere.com hosts its responsible-disclosure policy and (per the site) compliance documentation.

Customers: a logo wall and a smaller set of named case studies

Cohere's North page lists "organizations that trust us" including Oracle, Dell Technologies, RBC, LG CNS, Fujitsu, Bell, Asana, SAP, Salesforce, TD Bank, McKinsey & Company and Accenture -- a vendor-published logo wall, not independently confirmed beyond the logo itself. It backs a smaller subset with detailed, named case studies: CoreWeave (North cut customer-support resolution time within 90 days, per Cohere), Draftwise (a self-reported 30% lift in search-result quality after adopting Command, Embed and Rerank, per Draftwise's own internal benchmark), Fujitsu (co-developed the Takane Japanese-language model), Oracle (says it has launched over 100 generative-AI use cases on Cohere), BambooHR and Atomicwork.

Company, funding and research arm

Cohere says it raised nearly $1 billion across four funding rounds (Series A through D) between 2021 and 2024; its own about page does not disclose a newer round as of today. Cohere Labs, its open-science research arm launched in 2022, says it has 4,500+ community members and has published over 100 research papers. These are Cohere's own reported totals, not independently broken out by round or audited.

How much does Cohere cost?

Prices excluding tax, as Cohere publishes them.

  • North

    Custom

    • Interface
    • Purpose-built generative models
    • Intelligent search
    • AI agents for routine tasks and complex workflows
  • Compass

    Custom

    • Pre-built data connectors
    • Intelligent search
    • Document parsing
    • Managed index
Cohere's pricing page

Pricing page, captured

cohere.com/pricing, checked

All plans and limits

PlanPriceMain limits
CompassPrice on requestNone stated
Model Vault A100 40GB$3.25 per GPU hourNone stated
Model Vault A100 80GB$3.50 per GPU hourNone stated
Model Vault B200 192GB$9.75 per GPU hourNone stated
Model Vault B200 192GB Encrypted (Confidential Compute)$12 per GPU hourNone stated
Model Vault GB200 192GB$12.50 per GPU hourNone stated
Model Vault H100 80GB$6.25 per GPU hourNone stated
Model Vault H100 80GB Encrypted (Confidential Compute)$9 per GPU hourNone stated
Model Vault H200 141GB$7.25 per GPU hourNone stated
Model Vault L4 24GB$3 per GPU hourNone stated
Model Vault MI300X 192GB$6 per GPU hourNone stated
NorthPrice on requestNone stated
Trial API key$0 per API callUsage: rate limited and are not permitted to be used for production or commercial purposes

Read from cohere.com/pricing.

What we checked on Cohere

YesPartlyNo

What it costs to startYes

What the price counts

Usage-based across the board: generative models per token, Rerank per search, Embed per token embedded, while Model Vault bills per dedicated instance by the hour or month and the North and Compass platforms are quote only.

Our generative models, such as Command A+, Command A, Command R7B, Command R, and Command R+, are priced on a per-token basis. ... Our Rerank models…

docs.cohere.com/v2/docs/how-does-cohere…Checked

What it costs to startYes

Free to use

A free evaluation key covers every model and endpoint after signing up, capped at 1,000 API calls a month and low per-minute rates, so it serves prototypes rather than production traffic.

trial API key usage is free, but limited. Developers wanting to test different applications or build proofs of concept can use all of Cohere's models…

docs.cohere.com/v2/docs/how-does-cohere…Checked

How you build with itYes

Public API

Public documented REST API with v1 and v2 versions, covering chat, embed, rerank, parse, tokenize, embed jobs and audio transcription endpoints, with published rate limits per key type.

Our models can be accessed directly through our API, allowing for the creation of scalable production workloads.

docs.cohere.com/v2/docs/how-does-cohere…Checked

How you build with itPartly

MCP server

The documentation site publishes an MCP endpoint that coding assistants can connect to, but it serves Cohere's docs; no MCP server fronting the models or the North platform is documented.

For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.cohere.com/_mcp/server

docs.cohere.com/llms.txtChecked

How you build with itPartly

Open source

Research models and datasets from Cohere Labs, including the Aya and Tiny Aya families, are released openly on Hugging Face, while the Command models, the API and the North platform are proprietary commercial products.

Aya's models and datasets are openly licensed, fostering global collaboration and enabling researchers to audit, extend, and innovate responsibly.

cohere.com/research/ayaChecked

Cohere alternatives

Questions about Cohere

How much does Cohere cost?
Cohere's Trial API key plan is free. Its paid plans include Compass, Model Vault A100 40GB at $3.25 per GPU hour and Model Vault A100 80GB at $3.50 per GPU hour. From cohere.com/pricing, checked 12 November 2026.
Does Cohere have an API or an MCP server?
Cohere has an API. From docs.cohere.com/v2/docs/how-does-cohere-pricing-work, checked 1 November 2026. Cohere has a limited MCP server. From docs.cohere.com/llms.txt, checked 1 November 2026.
Is Cohere open source, and can you host it yourself?
Part of Cohere is open source. From cohere.com/research/aya, checked 1 November 2026.
What have we checked on Cohere?
5 facts, under these headings: What it costs to start, how you build with it.
How do we check Cohere?
We hold 5 facts on Cohere, quoted from 7 pages. Facts last checked 1 November 2026. Price last checked 12 November 2026. Last change found 1 April 2026.

How we check every listing

Cohere price history

Cohere had a free plan, and paid plans from $3 per instance hour, on both of our reads, from 1 November 2026 to 12 November 2026.

PricePlanFirst seenLast seenReadingsSource
$3 per instance hourCheapest paid plan1 November 202612 November 20262Pricing page, read 1 November 2026

From our research, read twice since 1 November 2026. A date is when we read the page, not when the price changed.

Similar to Cohere

Is Cohere yours?

Free, and the page then carries the verified owner mark. You confirm it with an email at cohere.com. Claim this listing

Report an error if a fact is wrong.