Join our Newsletter — 33% off our NHI Course
Home› Guides› Cohere Models: Every Model, Spec, Price and Release…
Guide AI Platforms

Cohere Models: Every Model, Spec, Price and Release Date

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 15 November 2026 17 min read
On this page

Cohere is an enterprise AI company whose Command, North, Aya, Embed, Rerank, Parse and Transcribe models are offered through the Cohere API, private deployments and partner clouds, and its line-up changed several times in 2026. This hub lists every model in Cohere's public documentation as of 15 November 2026, with model IDs, release dates, context windows, prices, availability, licences and retirement dates. It is written for developers, buyers, architects and anyone who needs to know which Cohere model is which. Models are listed by release date, newest first, an objective order that implies no ranking. Every fact comes from Cohere's own documentation, pricing page, changelog or Cohere Labs model cards, cited in the sources list.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. It uses Cohere's public documentation as of November 2026. Model details change, so confirm against Cohere's current documentation.

Key takeaways

  • Seven model families are live. Cohere's models page lists Command, North, Embed, Rerank, Parse, Transcribe and Aya models; the newest generative model in the Command line is Command A+ (command-a-plus-05-2026), announced on 20 May 2026.
  • Context runs from 8K to 256K tokens. Command A, Command A Reasoning and North Mini Code list 256K; Command A+, Command A Vision and the Command R models list 128K; Tiny Aya and Command A Translate list 8K.
  • Public token prices cover legacy models only. The pricing page lists per-token rates for Command, Command R and Aya Expanse, and per-GPU hourly rates for Model Vault; current Command, Embed 5 and Rerank 5 token prices are not stated on the public pricing page.
  • Several models have open weights. Cohere Labs publishes Command A+ under Apache 2.0, and Command A and Tiny Aya under CC-BY-NC 4.0; Cohere describes Cohere Transcribe as an open source model.
  • Retirements are dated. Embed v2 models and Aya Expanse 8B and Aya Vision 8B were retired on 4 April 2026, and the original Command, Command Light, Command R and Command R+ were deprecated effective 15 September 2025.

What is Cohere?

Cohere builds language, embedding, reranking, document parsing and speech models for enterprise use, and its research arm, Cohere Labs, publishes model weights on Hugging Face. Its documentation groups models by endpoint: Chat (Command, North, Aya), Embed, Rerank, Parse and Audio Transcriptions. Model IDs carry a month and year suffix, such as command-a-03-2025, or a version number, such as embed-v5.0-pro.

Cohere sells access through Trial and Production API keys, Model Vault dedicated GPU capacity and private deployment. The main table lists the generative and speech models; Embed and Rerank models follow with their own fields. Facts are as of 15 November 2026.

Current Cohere models at a glance

As of 2026-11-15. Source: Cohere models page, changelog and Cohere Labs model cards. Models are listed by release date, newest first; where the changelog gives no date, the month comes from the model ID. Category is the Cohere model family.

ModelAPI ID or versionCategoryContextLicenceStatus
North Small Translate 1.0north-small-translate-1.0North16KNot stated by CohereLive (evaluation on the Cohere API)
Parse v5.0parse-v5.0ParseNot stated by CohereNot stated by CohereLive
Cohere Transcribe Arabiccohere-transcribe-arabic-07-2026TranscribeNot stated by Cohere (25MB file limit)Not stated by CohereLive
North Mini Code 1.0north-mini-code-1.0North256KNot stated by CohereLive (evaluation; production via Model Vault)
Command A+command-a-plus-05-2026Command128KApache 2.0 (weights)Live
Cohere Transcribecohere-transcribe-03-2026TranscribeNot stated by Cohere (25MB file limit)Described by Cohere as open sourceLive
Tiny Aya (Global, Earth, Fire, Water)tiny-aya-global, tiny-aya-earth, tiny-aya-fire, tiny-aya-waterAya8KCC-BY-NC 4.0 (weights)Live
Command A Reasoningcommand-a-reasoning-08-2025Command256KNot stated by CohereLive
Command A Translatecommand-a-translate-08-2025Command8KNot stated by CohereLive
Command A Visioncommand-a-vision-07-2025Command128KNot stated by CohereLive
Command Acommand-a-03-2025Command256KCC-BY-NC 4.0 (weights)Live
Command R7Bcommand-r7b-12-2024Command128KNot stated by CohereLive
Command R and Command R+ (08-2024)command-r-08-2024, command-r-plus-08-2024Command128KNot stated by CohereLive
Aya Expanse 32B and Aya Vision 32Bc4ai-aya-expanse-32b, c4ai-aya-vision-32bAya128K and 16KNot stated by CohereLive

Command A+ and Command A Vision accept text and images; Aya Vision 32B accepts text and images; the other Chat models in the table are text in, text out. The Transcribe models take audio and return text. See Open-Weight vs Closed Models Explained.

Key facts for the main Cohere models

As of 2026-11-15. Identical fields for every model; "Not stated by Cohere" means not stated by Cohere on the pages checked (2026-11-15).

ModelReleasedMax outputKnowledge cutoffInput / outputOpen weightsParametersPositioning (provider wording)
North Small Translate 1.02026-09-0916KNot stated by CohereText / textNot stated by Cohere218B total, 25B active (MoE)Machine translation across 50+ languages
North Mini Code 1.02026-06-0964KNot stated by CohereText / textNot stated by Cohere30B total, 3B active (MoE)Agentic coding
Command A+2026-05-2064KNot stated by CohereText, image / textYes (Apache 2.0)218B total, 25B active (MoE)First Mixture of Experts model in the Command A family, adding vision, agentic, reasoning and translation
Tiny Aya Global2026 (ID undated)8KNot stated by CohereText / textYes (CC-BY-NC 4.0)3.35BBalanced coverage across regions, 70 languages
Command A Reasoning2025-08 (from ID)32KNot stated by CohereText / textNot stated by CohereNot stated by CohereCohere's first reasoning model, 23 languages
Command A Vision2025-07 (from ID)8KNot stated by CohereText, image / textNot stated by CohereNot stated by CohereCharts, tables, OCR and document question answering
Command A2025-03 (from ID)8KNot stated by CohereText / textYes (CC-BY-NC 4.0)111BTool use, agents, RAG and multilingual tasks; runs on two GPUs
Command R7B2024-12 (from ID)4KNot stated by CohereText / textNot stated by CohereNot stated by CohereSmall, fast model for RAG, tool use and agents

Embed and Rerank models

As of 2026-11-15. Source: Cohere models page and changelog. Max output tokens do not apply to these models.

Model IDReleasedModalitiesContextDimensions or notes
rerank-v5.0-pro, rerank-v5.0-fast2026-10-21Text32KMultilingual, handles JSON, Context Focus and calibrated scores
embed-v5.0-pro, embed-v5.0-fast2026-09-30Text, images, mixed (e.g. PDFs)128K256 to 2048 dimensions, default 2048
rerank-v4.0-pro, rerank-v4.0-fastNot stated by CohereText32KPrevious generation; v5.0 recommended for new projects
embed-v4.0Not stated by CohereText, images, mixed128K256 to 1536 dimensions, default 1536; Embed Jobs supported
rerank-v3.5, rerank-english-v3.0, rerank-multilingual-v3.0Not stated by CohereText4Kv3 series
embed-english-v3.0, embed-multilingual-v3.0 and light variantsNot stated by CohereText, images (as listed)5121024 or 384 dimensions, cosine

See the RAG Security Guide for the controls that apply when embedding and reranking models sit inside a retrieval pipeline.

Cohere model families and naming

  • Command: the main generative line, from Command R and R+ (2024) to Command A (March 2025), its Reasoning, Vision and Translate variants, and Command A+ (May 2026).
  • North: two models on the Cohere API for evaluation as of November 2026: North Mini Code 1.0 for agentic coding and North Small Translate 1.0 for translation across 50+ languages.
  • Aya: Cohere Labs multilingual models: Tiny Aya (3.35B, four regional variants, 70 languages), Aya Expanse 32B and Aya Vision 32B.
  • Embed and Rerank: retrieval models in pro and fast variants from version 4.0 onward; version 5.0 is the latest of each.
  • Parse: parse-v5.0, a vision model that extracts structured data from enterprise documents.
  • Transcribe: speech-to-text, with a general multilingual model and an Arabic fine-tune.
  • IDs: Chat model IDs end in a month and year (command-a-plus-05-2026 is May 2026); retrieval models use a version number. Some older IDs were aliases, for example command-r pointed to command-r-03-2024. See LLM Naming and Versioning Explained.

When was each Cohere model released?

As of 2026-11-15. Dates are from the Cohere changelog; earlier rows use the month in the model ID, as the changelog pages checked give no announcement date for them.

DateEventAPI ID
2024-08 (from ID)Command R and Command R+ August 2024 updatescommand-r-08-2024, command-r-plus-08-2024
2024-12 (from ID)Command R7Bcommand-r7b-12-2024
2025-03 (from ID)Command Acommand-a-03-2025
2025-07 and 2025-08 (from IDs)Command A Vision, Command A Translate and Command A Reasoningcommand-a-vision-07-2025, command-a-translate-08-2025, command-a-reasoning-08-2025
2025-09-15Original Command, Command Light, Command R and Command R+ deprecatedcommand, command-light, command-r-03-2024, command-r-plus-04-2024
2026-03 (from ID)Cohere Transcribecohere-transcribe-03-2026
2026-04-04Embed v2 and Aya 8B models retiredembed-english-v2.0, c4ai-aya-expanse-8b, c4ai-aya-vision-8b and others
2026-05-20Command A+ announcedcommand-a-plus-05-2026
2026-06-09North Mini Code announcednorth-mini-code-1.0
2026-07-07Cohere Transcribe Arabic announcedcohere-transcribe-arabic-07-2026
2026-08-27Parse announcedparse-v5.0
2026-09-09North Small Translate announcednorth-small-translate-1.0
2026-09-30Embed 5 announcedembed-v5.0-pro, embed-v5.0-fast
2026-10-21Rerank 5 announcedrerank-v5.0-pro, rerank-v5.0-fast
2026-11-03Open models in Model Vault; Model Vault Encrypted describedNot applicable
2026-11-04Embed 5 and Rerank 5 available on OpenRouterembed-v5.0-pro, rerank-v5.0-pro and fast variants

The latest model release is Rerank 5, announced on 21 October 2026. The latest Command model is Command A+ (20 May 2026).

How much do Cohere models cost?

As of 2026-11-15. USD per million tokens (MTok) for token-priced models. Source: Cohere pricing page. The public pricing page lists per-token rates only for legacy models for existing customers and Aya Expanse; "Not stated" means not stated by Cohere.

ModelInputCached inputOutputBatchCache writeNotes
Command R+ 08-2024$2.50Not stated$10.00Not statedNot statedLegacy, existing customers
Command R+ 04-2024$3.00Not stated$15.00Not statedNot statedLegacy, existing customers
Command R 03-2024$0.50Not stated$1.50Not statedNot statedLegacy, existing customers
Command$1.00Not stated$2.00Not statedNot statedLegacy, existing customers
Command Light$0.30Not stated$0.60Not statedNot statedLegacy, existing customers
Aya Expanse (8B and 32B)$0.50Not stated$1.50Not statedNot stated8B retired 2026-04-04
Command A+, Command A family, North, Embed 5, Rerank 5Not stated by CohereNot on the public pricing page checked (2026-11-15)

Model Vault pricing

As of 2026-11-15. USD per GPU per hour. The pricing page shows an on-demand price and a promotional price valid through 31 March 2027; pricing is per GPU and replica, and reserved or committed capacity is through sales.

GPUOn-demandPromotional (to 2027-03-31)
L4 24GB$3.00$1.00
A100 40GB$3.25$2.25
A100 80GB$3.50$2.25
MI300X 192GB$6.00$4.00
H100 80GB$6.25$4.25
H100 80GB (Encrypted)$9.00$4.25
H200 141GB$7.25$5.50
B200 192GB$9.75$8.00
B200 192GB (Encrypted)$12.00$8.00
GB200 192GB$12.50$10.00
  • Rerank billing unit: one search unit is one query with up to 100 documents; documents over 500 tokens, counting the query, are split into chunks and each chunk counts as a document.
  • Trial keys: free and rate limited, not permitted for production or commercial use.
  • Billing: invoices are issued at the end of each calendar month or when the outstanding balance reaches $250.
  • OpenRouter: Embed 5 and Rerank 5 prices on OpenRouter are listed on each model's OpenRouter page, per the 4 November 2026 changelog entry.

See Model Pricing Explained, the LLM Pricing Tracker and How to set spend limits on LLM API keys.

Where are Cohere models available?

  • First-party: the Cohere API (Chat, Embed, Rerank, Parse and Audio Transcriptions endpoints), with Trial and Production keys.
  • Model Vault: dedicated GPU capacity priced per GPU hour, with an Encrypted variant that uses confidential computing; North Mini Code supports production deployment through Model Vault, and open models were added to Model Vault on 3 November 2026.
  • Cloud platforms: Cohere's platform support page lists Amazon Bedrock, Amazon SageMaker, Azure and Oracle OCI, each supporting a subset of features. Chat, Embed and Rerank are listed for Bedrock, SageMaker and Azure; Chat and Embed for OCI.
  • Private deployment: on-premises or isolated VPC deployments, with all SDK features listed as supported.
  • Other channels: Cohere Transcribe is on Microsoft Foundry as coherelabs-cohere-transcribe-03-2026; North Mini Code is on OpenRouter as cohere/north-mini-code:free; Embed 5 and Rerank 5 are on OpenRouter.
  • Open weights: Command A+, Command A and Tiny Aya weights are on Hugging Face under Cohere Labs.

Confirm the exact model ID in the chosen cloud's catalogue and region.

Restricted-access Cohere models

As of 2026-11-15. Source: Cohere models page, pricing page and deprecations page.

ModelAccess stated by providerData retention condition
North Small Translate 1.0Available on the Cohere API for evaluationNot stated by Cohere
North Mini Code 1.0Available on the Cohere API for evaluation; production deployment through Model VaultModel Vault: zero data retention by default
Legacy Command and Command R 03/04-2024 modelsDeprecated: unavailable to new customers; existing users (use within 90 days of the announcement) keep access until shutdownNot stated by Cohere

Cohere's documentation names no model that is limited to a verified or invited group of organisations.

What is the context window of Cohere models?

Command A, Command A Reasoning and North Mini Code list 256K-token context; Command A+, Command A Vision, Command R7B, the Command R models, Aya Expanse 32B and the Embed v4 and v5 models list 128K. North Small Translate lists 16K, Aya Vision 32B 16K, and Tiny Aya and Command A Translate 8K. Rerank v4 and v5 accept 32K and Rerank v3 4K. Max output is 64K on Command A+ and North Mini Code, 32K on Command A Reasoning, 8K on Command A and its Vision and Translate variants, and 4K on Command R models. The Command A model card notes that the model supports 256K but is configured in Hugging Face for 128K. See Context Windows Explained.

Is any Cohere model open source?

Some are. The Cohere Labs Hugging Face cards publish Command A+ weights under Apache 2.0, and Command A (111B, described as an open weights research release) and Tiny Aya under CC-BY-NC 4.0, a non-commercial licence; the CC-BY-NC cards also require adherence to the Cohere Labs Acceptable Use Policy. Cohere's models page describes Cohere Transcribe as an open source model. Licences for the other models are not stated on the pages checked. See the Open-Weight Model Licence Tracker.

Model cards and system cards

As of 2026-11-15.

ModelModel card or pageSystem cardTechnical report
Command A+Cohere Labs Hugging Face card (architecture, languages, licence, hardware)Not stated by Cohere on the pages checked (2026-11-15)Not stated by Cohere on the pages checked
Command ACohere Labs Hugging Face cardNot stated by Cohere on the pages checked (2026-11-15)Paper linked from the card (published 1 April 2025)
Tiny AyaCohere Labs Hugging Face cardNot stated by Cohere on the pages checked (2026-11-15)Paper "Tiny Aya: Bridging Scale and Multilingual Depth" linked from the card
Other models listed aboveModels page entry (ID, context, max output, modality, endpoint)Not stated by Cohere on the pages checked (2026-11-15)Not stated by Cohere

See Model Cards and System Cards Explained.

Benchmarks

As of 2026-11-15. Provider-reported and independent results are kept in separate columns.

ModelProvider-reported benchmarksIndependent leaderboard results
Command A+Not stated on the model card (card refers to the Cohere blog)None cited on this page
Command AGPQA Diamond 50.51 (Cohere Labs model card)None cited on this page
Tiny Aya GlobalGPQA Diamond 28.28 (Cohere Labs model card)None cited on this page
North, Embed 5, Rerank 5Not stated on the pages checkedNone cited on this page

See How to Read LLM Benchmarks.

Deprecations and retirements

As of 2026-11-15. Source: Cohere deprecations page and changelog. Cohere's lifecycle stages are Active, Legacy, Deprecated and Shutdown; impacted customers are notified by email, documentation and blog posts, and the page states no minimum notice period in days.

ModelAPI IDDeprecated or noticeRetirement or shutdownStated replacement
Embed English v2, Light v2, Multilingual v2embed-english-v2.0, embed-english-light-v2.0, embed-multilingual-v2.0Not stated by Cohere2026-04-04 (retired)embed-english-v3.0, embed-multilingual-v3.0 or embed-v4.0
Aya Expanse 8B, Aya Vision 8Bc4ai-aya-expanse-8b, c4ai-aya-vision-8bNot stated by Cohere2026-04-04 (retired)command-r7b-12-2024, command-a-03-2025 or command-a-reasoning-08-2025
Command R (03-2024), Command R+ (04-2024) and aliasescommand-r-03-2024, command-r-plus-04-2024, command-r, command-r-plusEffective 2025-09-15Not stated by Coherecommand-r-08-2024, command-r-plus-08-2024 or command-a-03-2025
Command, Command Lightcommand, command-lightEffective 2025-09-15Not stated by Coherecommand-r-08-2024, command-r-plus-08-2024 or command-a-03-2025
Rerank v2rerank-english-v2.0, rerank-multilingual-v2.02024-12-022025-04-30 (retired)rerank-v3.5
Rerank v4rerank-v4.0-pro, rerank-v4.0-fastN/A (live, previous generation)Not stated by Coherererank-v5.0-pro or rerank-v5.0-fast recommended for new projects

Endpoints and features deprecated effective 15 September 2025 include /v1/generate, /v1/summarize, /v1/classify and /v1/connectors; fine-tuning for older models was retired on the same date. See the AI Model Retirement Guide.

Data handling and security

  • Training use: Cohere's enterprise data commitments page states that enterprise customers can opt out in the dashboard data settings, after which prompts and generations are not used to train Cohere models; the opt-out applies to data created after the change.
  • Default retention: on the SaaS platform, logged prompts and generations are deleted after 30 days, except where retention is required by law or contract or usage is flagged as violating Cohere's terms.
  • Zero data retention (ZDR): offered to enterprise customers who make additional usage commitments, on request to Cohere support; Cohere still receives usage data. Model Vault has zero data retention by default, per the 3 November 2026 changelog.
  • Private and partner deployments: for on-premises, isolated VPC and third-party cloud deployments, Cohere states it does not access or process customer prompts or generations.
  • Confidential compute: Model Vault Encrypted uses confidential computing so that neither Cohere nor the cloud provider can see prompts or responses.
  • Key handling: generic controls apply to every provider's API: keep keys out of code, scope and rotate them, and set spend limits.

See AI Data Retention and Deletion and the Open-Weight Model Security Guide for self-hosted weights.

How to choose a Cohere model: factual criteria

  • Task and endpoint: Chat models for generation, Embed and Rerank for retrieval, Parse for document extraction, Transcribe for audio.
  • Modality: image input is listed for Command A+, Command A Vision and Aya Vision 32B.
  • Context and output: 256K context on Command A, Command A Reasoning and North Mini Code; 64K output on Command A+ and North Mini Code.
  • Licence: Apache 2.0 for Command A+ weights; CC-BY-NC 4.0 (non-commercial) for Command A and Tiny Aya weights.
  • Pricing route: confirm token prices for current models with Cohere, or compare Model Vault GPU hours, noting the promotional period ends 31 March 2027.
  • Retention: SaaS logs are kept 30 days by default; ZDR, Model Vault and private deployment change that.
  • Lifecycle: avoid deprecated IDs and check the Legacy and Deprecated stages before building.
  • Platform: confirm the model ID and feature set on Bedrock, SageMaker, Azure or OCI, as each supports a subset.

Practitioner checklist

  • Pin production code to a dated model ID, not an alias, and record it in the AI inventory.
  • Replace any remaining calls to command, command-light, command-r or command-r-plus aliases and to /v1/generate or /v1/summarize.
  • Confirm that no workload still calls the Embed v2 or Aya 8B IDs retired on 4 April 2026.
  • Use Production keys for commercial workloads; Trial keys are not permitted for production use.
  • Set the training opt-out in the dashboard data settings, or request ZDR, before sending sensitive data.
  • Check the licence on the Hugging Face card before self-hosting Command A or Tiny Aya commercially.
  • Store Cohere API keys in a secrets manager and set spend limits. See the LLMjacking Guide.

Standards and references

Related NHI Mgmt Group resources: Cohere Command A family, LLM Release Tracker, Reasoning Models Explained, LLM Security Guide.

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 15 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.