Tech & AI, Explained

cohere

10 min read

🇨🇦 Canada · Cohere

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Canada’s frontier AI lab — enterprise-focused, multilingual-first, deeply rooted in the Transformer-architecture lineage (one of its founders co-wrote “Attention Is All You Need”). The encyclopedia’s recommended pick when you want a non-US, non-Chinese, enterprise-grade AI vendor.

Front-matter facts

FieldValue
VendorCohere Inc (Toronto, Canada) — with offices in San Francisco, London, New York
Country / origin🇨🇦 Canada
Recommended for Australian users?✅ Yes — fully accessible from AUS; strong enterprise posture; Canadian privacy framework (PIPEDA) generally well-aligned with AUS Privacy Act
Privacy summaryAPI + Enterprise: never trains on your data by default. Cohere has historically been API-first; the consumer “Coral” chat surface (now “North”) has clearer enterprise no-training defaults than competitors.
Free tierYes — free API trial keys with rate limits; North (the consumer chat) free tier available
Paid tiersAPI pay-per-token (Command R, Command R+, Command A, plus Embed, Rerank) · North (enterprise chat) quoted · Private deployment quoted
First releasedCohere founded 2019; Command models 2022; Coral chat 2023; rebranded “North” 2024
Last reviewed2026-06-26
Official sitehttps://cohere.com

What it is

Cohere is a Canadian frontier AI lab founded in 2019 by Aidan Gomez (a co-author of the 2017 paper “Attention Is All You Need” — the paper that introduced the Transformer architecture, which every modern LLM uses), Nick Frosst, and Ivan Zhang. Headquartered in Toronto.

Unlike OpenAI, Anthropic, or Google — which all started with a strong consumer chat product — Cohere has been explicitly enterprise-first. Its flagship products are:

  • Command family of models (Command R, Command R+, Command A) — designed for enterprise RAG (retrieval-augmented generation), agent workflows, and multilingual content
  • Embed — high-quality text embedding models for search
  • Rerank — a search-result reranking model that meaningfully improves any RAG system
  • North — Cohere’s enterprise chat / workspace product (formerly “Coral”); positioned as a privacy-respecting Microsoft Copilot / Google Workspace AI alternative
  • Private deployment — Cohere will deploy its models inside your VPC / your own datacentre (rare among frontier labs; meaningful for highly regulated industries)

Cohere also released Aya 23 / Aya Expanse — research-grade multilingual models specifically tuned for languages underrepresented in other frontier models.

Analogy: if Claude / ChatGPT / Gemini are “consumer-first labs that also sell to enterprise,” Cohere is “the enterprise-first lab that also has a consumer surface.” It’s the one Australian banks, hospitals, and government agencies are most likely to pick when they want a non-US, non-Chinese frontier AI vendor with deep enterprise controls.

What you’d use it for

As a developer

  • RAG / search — Cohere’s Embed + Rerank are widely considered best-in-class for RAG pipelines
  • Multilingual content generation — Command and Aya models handle non-English languages exceptionally well
  • Enterprise agent workflows — Command R+ is specifically tuned for tool use and grounded responses
  • Long-context summarisation — Command R+ handles 128K tokens
  • Citation-grounded responses — Command models natively produce inline citations to source documents in RAG mode

As a business user (via North)

  • Document Q&A and search across your company’s knowledge base
  • Workspace integrations (Notion, Slack, Google Drive, Confluence, etc.)
  • Multilingual customer-support agents
  • Compliance-friendly chat that’s never been trained on your data

What you’d NOT use Cohere for

  • General consumer chat as a personal AI buddy — Claude / ChatGPT / Gemini have far more polished consumer surfaces
  • Image generation — Cohere doesn’t make image models
  • Video / music / voice — not part of Cohere’s portfolio
  • Casual coding help — capable, but Claude Code / Cursor / Copilot have much more developed coding-tool UX

How to sign up + first 5 minutes from Australia

Free API trial (developer path)

  1. Go to dashboard.cohere.com. Sign up with email or Google.
  2. Verify email; AUS mobile (+61) accepted for any phone verification.
  3. Free trial API key is generated immediately. Limited rate (good enough for prototyping); no credit card required.
  4. Try it in the Playground (dashboard.cohere.com/playground) — pick Command R or Command R+, type a prompt, see a response.
  5. For production: add credit card → switch to paid keys (higher rate limits).

North (enterprise chat) free trial

  1. Go to north.cohere.com. Sign up with your work email.
  2. North free tier: small daily usage allowance; documents you upload aren’t used for training.
  3. For full use: contact Cohere sales for a paid plan (per-seat pricing varies by org size and feature set; expect ~US$30-50/user/month range).

Via AWS Bedrock / Oracle Cloud / Azure / Cohere on Google Cloud

  • Cohere models are available via several Western cloud providers — useful for AUS data residency
  • AWS Bedrock Sydney (ap-southeast-2) hosts Cohere Command models — AUD billing through AWS, AUS data residency
  • Oracle Cloud has a long-standing Cohere partnership

For the Cohere Playground (no commit)

  • Go to coral.cohere.com or dashboard.cohere.com/playground — instant access with sign-up
  • No production-grade SLAs; good for testing

What it costs — what you actually get

Free trial API

  • Free API key with rate-limited Command R, Command R+, Embed, Rerank access
  • Sufficient for prototyping and learning
  • No credit card required

Production API (pay-per-token)

  • Command R+ (flagship): ~US10 per million output tokens
  • Command R: ~US0.60 per million output tokens (much cheaper)
  • Command A (newest, multimodal): pricing varies
  • Embed (embedding model): ~US$0.10 per million tokens
  • Rerank: per-search pricing
  • All prices subject to change — verify at cohere.com/pricing

North (enterprise chat)

  • Per-seat pricing quoted
  • Includes: document Q&A, workspace integrations, multilingual support, contractual no-training, admin controls
  • Free trial available

Private deployment

  • Cohere will deploy models inside your VPC (AWS, Azure, GCP, Oracle, on-prem)
  • Quoted pricing — typically used by regulated enterprises (banks, healthcare, government)
  • Same models as the public API but with full data isolation

Via AWS Bedrock

  • Command R / Command R+ available in Bedrock at AWS-billing rates
  • AUS data residency via Sydney region (ap-southeast-2)
  • No separate Cohere account needed

Hidden costs to know about

  • Cohere is intentionally enterprise-focused — there’s no Pro consumer subscription like Claude Pro or ChatGPT Plus. Free trial → paid API or North.
  • Pricing on Command R is dramatically cheaper than Command R+ — pick the right model for the task to control cost.
  • North per-seat pricing typically requires multi-seat commitments.

How it compares to alternatives

CapabilityCohereAnthropic ClaudeOpenAI GPTGoogle Gemini
Consumer chat polishLess developed (North)Best (Claude.ai)Best (ChatGPT)Best (Gemini)
Enterprise focusMost enterprise-firstStrongStrongStrong
MultilingualStrongest on non-English (Aya, Command)StrongStrongStrong
RAG-specific toolingBest — Embed + Rerank + grounded CommandGoodGoodGood (Vertex AI Search)
Long context128K (Command R+)1M (Sonnet)256K1M (Gemini 3 Pro)
Image generationNoneNone nativegpt-image-1Imagen
Video generationNoneNoneSoraVeo 3
VoiceNoneRolling outBest (Advanced Voice)Good (Gemini Live)
CodingCapableBest (Claude Code)StrongStrong
Private deployment in your VPCYes (rare)Limited (via Bedrock)Limited (via Azure)Limited (via Vertex AI)
Canadian / non-US sovereigntyYes (Canadian)US-basedUS-basedUS-based

When to pick Cohere:

  • You’re building enterprise RAG / search (Embed + Rerank are genuinely best-in-class)
  • You need multilingual generation (Aya / Command beat alternatives for low-resource languages)
  • You want a non-US frontier vendor without going European
  • You need private VPC deployment of frontier-quality models
  • You’re an Australian organisation with Canadian connections or preferences

When NOT to pick Cohere:

  • You want polished consumer chat — use Claude / ChatGPT / Gemini
  • You need image / video / voice — Cohere doesn’t make those
  • You’re a casual personal user — Cohere’s free trial is developer-focused

Privacy / data handling deep dive

API + North + Enterprise: no training by default

Cohere’s API and enterprise products do NOT train on your inputs by default. This is contractually committed in the Cohere Terms of Service and the North Data Processing Addendum. Cohere has been API-first since founding, which means its privacy posture is enterprise-grade from the start — unlike consumer-first labs that bolted on enterprise terms later.

Free trial

Free trial API keys may have slightly weaker terms — verify the current dashboard.cohere.com/account/settings privacy controls. The trial is intended for development; don’t paste production data into it.

Data retention

API inputs retained briefly for abuse-monitoring (typically 30 days) then deleted. North conversations retained per your tenant’s admin-configured retention policy.

Where data lives

Cohere has data centres in the US (primary) and Canada. Enterprise customers can specify region — including via:

  • AWS Bedrock Sydney (ap-southeast-2) for AUS data residency
  • Oracle Cloud Toronto / Sydney regions
  • Google Cloud / Azure regions where Cohere is hosted
  • Private VPC deployment — your data stays in your infrastructure entirely

Canadian / international compliance

  • PIPEDA (Canadian Personal Information Protection and Electronic Documents Act) — natively compliant
  • GDPR — Data Processing Addendum available
  • SOC 2 Type II — audited
  • HIPAA — available via private deployment
  • Australian Privacy Act — Cohere is subject when handling AUS personal information; their PIPEDA-native posture generally meets APP requirements

For Australian government / regulated industries

Cohere is one of the few non-US frontier-AI vendors with enterprise maturity to be a viable supplier to AUS government, banking, healthcare, and other regulated sectors. The combination of:

  • Non-US sovereignty (Canada is part of AUKUS and Five Eyes, but not American jurisdiction)
  • Private VPC deployment option
  • PIPEDA-native compliance
  • Standard SOC 2 / HIPAA audits

…makes Cohere a recurring choice for AUS organisations evaluating alternatives to American AI providers.

See privacy-and-data-training.md 🟩 🟦 for the comparative deep dive.

Recent changes (since the last review)

  • 2026: Command A (next-generation multimodal) released
  • 2025: North launched (rebrand of Coral); positioned as Microsoft Copilot / Workspace AI alternative
  • 2025: Aya Expanse multilingual research models released
  • 2024: Command R and Command R+ launched — RAG-first design
  • 2024: Coral chat surface launched (later renamed North)
  • 2024: Embed v3 + Rerank v3 (the search-specific models that anchor Cohere’s RAG positioning)
  • 2023: Cohere reached US$5.5B valuation in funding round led by NVIDIA, AMD, Salesforce, Oracle
  • 2019: Company founded

(Check cohere.com/blog for the latest.)

Gotchas

  • Cohere is not a consumer-chat-first product. If you sign up expecting a ChatGPT competitor, you’ll find the experience less polished. Coral / North is fine; ChatGPT / Claude / Gemini consumer surfaces are simply more developed.
  • The model names are confusing — Command (legacy), Command R, Command R+, Command A, plus Aya research models, plus Embed, plus Rerank. For most users: Command R for cheap-and-fast, Command R+ for higher quality, Embed + Rerank for RAG.
  • Cohere is enterprise-priced. No “Pro tier for US$20/month.” The on-ramp is the free API trial → developer pay-per-token, or enterprise sales for North.
  • Aidan Gomez (co-founder/CEO) frequently does AI podcasts and conference talks — useful primary source for thinking about Cohere’s strategy.
  • Cohere doesn’t make image / video / music / voice models. If you need those capabilities alongside chat, plan to combine Cohere with another vendor (Imagen, Veo, Suno, ElevenLabs).
  • The private VPC deployment option is genuinely rare — most other frontier labs require you to send data to their cloud. If your compliance posture requires data-never-leaves-our-VPC, Cohere is often the only realistic frontier-AI vendor.
  • Cohere has investors / partners across the major Western cloud providers (NVIDIA, AMD, Salesforce, Oracle, Microsoft, Google, AWS). You’ll find Cohere models accessible through nearly all of them — pick the one that matches your existing cloud relationship for billing simplicity.
  • AWS Bedrock Sydney does not host all Cohere models — verify which models are available in your target region before committing.
  • Cohere occasionally appears in news around “the Canadian alternative” framing — this is a real positioning, not just marketing. The Government of Canada has explicitly invested in Cohere as part of national AI sovereignty strategy.

See also

Sources

Graph View

Backlinks