Enterprise-first LLM platform. Command A+ (released May 20, 2026) is now the agentic flagship: a 218B-parameter sparse MoE with 25B active parameters, 128K input context, 64K max generation, vision input, reasoning, tool use, and 48-language support under Apache 2.0. North Mini Code coding model with 256K total context, 64K max generation, Apache 2.0 weights, and availability through Hugging Face, Cohere API, OpenRouter, and Model Vault. Command R+ handles production RAG workloads at $2.50 input / $10 output per million tokens. Command R7B runs locally for edge cases. Embed v4 and Rerank 4 power one of the strongest dedicated retrieval stacks in the industry.
No consumer chat app. No image or video generation. Cohere sells to enterprise buyers: regulated industries, multilingual organizations, and teams that need private VPC, on-prem, or sovereign deployment. The official Cohere and Aleph Alpha announcement turns that sovereignty angle from a side note into one of the main reasons to consider Cohere in Europe.
On April 24, 2026, Cohere and Aleph Alpha announced a planned sovereign-AI combination. Cohere says the deal would combine Cohere’s global AI scale with Aleph Alpha’s European research and institutional relationships, while Schwarz Group intends to back the Series E with a EUR 500M structured financing commitment. The deal does not change public API pricing, but it strengthens Cohere’s European enterprise and sovereign-AI narrative.
Related coverage: Cohere releases Command A+ as an Apache 2.0 enterprise agent model, AI Industry Roundup, April 24, and AI News Desk, April 25.
System Verdict
Pick Cohere if you build production RAG, need multilingual retrieval, require a VPC / on-prem / sovereign deployment, or want open Apache 2.0 model control for enterprise agents and coding-model experiments. Embed v4 and Rerank 4 are among the best dedicated retrieval models shipping. Command A+ adds an Apache 2.0 flagship lane for agentic workflows, North Mini Code adds a developer-model lane, and North plus Model Vault support governed enterprise deployment. Model Vault runs the stack on isolated infrastructure, and the Cohere/Aleph Alpha move strengthens Cohere’s European enterprise story.
Skip it if you want a consumer chat app, image or video generation, an IDE assistant, or a broad plugin ecosystem. Cohere has no consumer UI competitive with ChatGPT or Claude. No image gen, no video. Smaller developer community than OpenAI or Anthropic. Command A+ still needs workload-specific testing against Claude Opus 4.8 and ChatGPT on peak reasoning tasks. North Mini Code is a model to evaluate inside a coding harness, not a finished Cursor or Copilot replacement.
Who pays which tier: Trial API key free for prototyping, Production API pay-as-you-go for most teams, Enterprise custom pricing for bulk workloads or private deployment, Model Vault dedicated instances ($2,500-$3,250/mo) for teams who need Embed or Rerank running on isolated hardware.
Key Facts
| Flagship model | Command A+ (command-a-plus-05-2026) · 218B sparse MoE · 25B active · Apache 2.0 · 128K input / 64K output |
| Production RAG model | Command R+ (command-r-plus-08-2024) · 128K context |
| Lightweight model | Command R7B (command-r7b-12-2024) · 7B params · 128K context |
| Embedding model | Embed v4 · multimodal · 256/512/1024/1536 dims · 100+ languages |
| Reranker | Rerank 4 (Fast + Pro variants) · 32K context · 100+ languages |
| Open models | Command A+ under Apache 2.0; Tiny Aya (3.35B params) supports 70+ languages |
| Developer model | North Mini Code · 30B total / 3B active MoE · Apache 2.0 · 256K total context / 64K output |
| Speech-to-text | Cohere Transcribe (transcribe-03-2026) · 14 languages · open-weights, edge-deployable |
| Enterprise agent platform | North (GA since early 2025) · private deployment via Model Vault |
| Private deployment | Model Vault (September 2025) · VPC + on-prem · isolated inference |
| API pricing (R+) | $2.50 input / $10 output per 1M tokens · Enterprise from $1 / $3 |
| Consumer chat | None |
| Image / video gen | None |
| Recent business move | Cohere and Aleph Alpha announced a planned sovereign-AI combination on April 24, 2026 |
Every data point above was verified against vendor sources and re-checked on June 18, 2026. See Sources.
Recent changes
- June 9, 2026: Cohere launched North Mini Code, its first agentic coding model. It is a 30B total / 3B active MoE with 256K total context, 64K max generation, Apache 2.0 weights, Hugging Face/API/OpenRouter/Model Vault availability, and a stated focus on coding agents, terminal tasks, code generation, and software-engineering workflows. This adds a developer-model evaluation lane to a page that previously read mostly as enterprise RAG plus Command A+.
- May 20, 2026: Cohere released Command A+, a 218B sparse MoE open under Apache 2.0 with 25B active parameters, 128K input context, 64K max generation, vision input, tool use, reasoning, and 48-language support. This materially strengthens Cohere’s sovereign/open enterprise model story.
- May 2026: Cohere Transcribe (transcribe-03-2026) listed as a new enterprise speech-to-text model, 14 languages, open-weights and edge-deployable, available via Hugging Face and the Cohere API. Adds an audio-input rail to the existing Command + Embed + Rerank stack.
- April 24, 2026: Cohere and Aleph Alpha announced a planned sovereign-AI combination backed by a Schwarz Group financing commitment. Still a sovereignty and enterprise-positioning story rather than a public API price change.
- April-June 2026: Pricing page keeps Command R+ 08-2024 at $2.50 input / $10 output per million tokens and Command R 08-2024 at $0.50 / $1.50. The buyer caveat is billing mechanics: a Rerank search covers one query and up to 100 documents, but documents over 500 tokens are automatically split into chunks that count toward ranked-document totals.
- June 2026 model lifecycle note: Current model docs list old Command, Command Light, Command R 03-2024, and Command R+ 04-2024 IDs as deprecated. Do not price a new Cohere deployment around those legacy IDs unless the account explicitly still uses them.
What it actually is
platform covering generation (Command family), coding-model evaluation (North Mini Code), retrieval (Embed v4), ranking (Rerank 4), transcription (Cohere Transcribe), and a private agent workspace (North). Cohere does not ship a consumer chat app. Buyers are compliance-aware enterprises, not retail users.
Command A+ is the agentic flagship: 128K input context, 64K max generation, sparse MoE architecture, vision input, tool use, reasoning, and 48-language support under Apache 2.0. North Mini Code is the developer-facing model: smaller, open, coding-agent-oriented, and designed for sovereign developer infrastructure rather than a bundled IDE. Command R+ remains the production RAG workhorse at a lower price point. Command R7B runs on modest hardware for on-device or edge inference.
The retrieval stack is the real moat. Embed v4 ships as a multimodal embedding model with Matryoshka dimensions (256, 512, 1024, 1536) across 100+ languages. Rerank 4 extends context to 32K tokens and leads public benchmarks on cross-lingual retrieval. No OpenAI or Anthropic product matches Cohere’s dedicated retrieval stack; the Aleph Alpha report adds a second moat around European enterprise relationships and sovereign-AI deployment credibility.
Model Vault, launched September 2025, runs Command, Embed, and Rerank inside isolated VPCs or on-prem infrastructure. Data never leaves the customer network. This is the deployment model regulated buyers actually sign.
Tiny Aya, released by Cohere Labs in February 2026, ships 3.35B-parameter open-weight multilingual models supporting 70+ languages on consumer hardware. Free to self-host.
When to pick Cohere
- You build production RAG and retrieval accuracy is the bottleneck. Embed v4 and Rerank 4 outperform OpenAI embeddings on cross-lingual and long-document retrieval benchmarks.
- You operate in many languages. Command A+ supports 48 languages, while Aya remains Cohere’s open multilingual research lane.
- You need VPC, on-prem, or sovereign-cloud deployment. Model Vault puts the full Command + Embed + Rerank stack inside your infrastructure. Data never crosses Cohere’s network.
- You sell into Europe or regulated markets. The reported Aleph Alpha combination gives Cohere a stronger European enterprise and sovereign-AI story.
- You build enterprise agents on internal data. North gives HR, finance, customer support, and IT teams a workspace for deploying Cohere-powered agents against proprietary documents.
- You want to test open coding-agent infrastructure. North Mini Code is a better fit for teams evaluating model control inside OpenCode, local harnesses, or Model Vault than for buyers looking for a finished AI IDE.
- Your compliance team already approves Oracle, AWS, or Azure. Cohere partners across those marketplaces and ships with SOC 2 and related controls pre-negotiated.
When to pick something else
- Consumer chat or image gen: ChatGPT. Largest ecosystem, GPT Image 2, Codex agent, custom-GPT marketplace.
- Peak reasoning or finished coding workflows: Claude, Claude Code, Cursor, or GitHub Copilot. North Mini Code is model infrastructure, not a finished coding assistant.
- Google Workspace integration: Gemini. Native Docs, Sheets, Gmail hooks Cohere cannot match.
- Open-weight self-hosting outside Cohere’s stack: Llama, DeepSeek, or Mistral AI. Command A+ is now Apache 2.0, but teams should still compare inference cost, ecosystem support, and deployment maturity.
- Solo dev or hobbyist use: ChatGPT Plus or Claude Pro at $20/mo. Cohere has no consumer subscription.
Pricing
API pricing via cohere.com/pricing, verified 2026-06-18:
| Plan / Model | Input ($/1M tok) | Output ($/1M tok) | Context | Who’s it for |
|---|---|---|---|---|
| Trial API key | Free | Free | Full | Prototyping, rate-limited |
| Command R+ 08-2024 | $2.50 | $10.00 | 128K | Production RAG workloads |
| Command R+ enterprise | $1.00 | $3.00 | 128K | Bulk or long-term contracts |
| Command R 08-2024 | $0.50 | $1.50 | 128K | Cost-sensitive chat + RAG |
| Command 06-2024 (legacy) | $1.00 | $2.00 | 4K | Existing customers only |
| Command Light (legacy) | $0.30 | $0.60 | 4K | Existing customers only |
| Embed v4 | Pay-per-token | n/a | 128K | Document + multimodal indexing |
| Rerank 4 | Pay-per-query | n/a | 32K | pipeline reranking |
Model Vault dedicated instances (for isolated VPC or on-prem inference):
| Model | Tier | Hourly | Monthly |
|---|---|---|---|
| Embed v4 | Small | $4.00 | $2,500 |
| Embed v4 | Medium | $5.00 | $3,250 |
| Rerank 3.5 | Medium | $5.00 | $3,250 |
| Rerank 4 Fast | Medium | $5.00 | $3,250 |
| Rerank 4 Pro | Medium | $5.00 | $3,250 |
| Rerank 4 Pro | Large | $10.00 | $6,500 |
Command A+ and North Mini Code managed pricing are not listed clearly on the public pricing page; contact sales, check the API dashboard, or verify the exact provider route before projecting production cost. Enterprise pricing on Command R+ drops input to $1 and output to $3 per 1M tokens, a roughly 60-70% discount from list. Public pricing rows last verified 2026-06-18 via Cohere pricing and Cohere pricing docs; the Command A+ and North Mini Code releases were re-verified 2026-06-18.
Against the alternatives
| Cohere (Command A+ + Embed v4) | OpenAI Enterprise | Anthropic Enterprise | |
|---|---|---|---|
| Context window | 128K input / 64K output (Command A+) | Undisclosed (GPT-5) | 1M (Opus / Sonnet) |
| Private deployment | Model Vault · VPC · on-prem | Azure OpenAI VPC | AWS Bedrock VPC |
| Dedicated retrieval stack | Embed v4 + Rerank 4 (strongest) | OpenAI embeddings (basic) | None (use third-party) |
| Multilingual | 100+ languages, strongest in retrieval-heavy multilingual evals | Strong | Strong |
| Consumer app | None | ChatGPT | Claude.ai |
| Image / video gen | None | GPT Image 2 | None |
| Agent platform | North workspace | Operator / GPT Store | Claude Code CLI |
| Developer model lane | North Mini Code, open Apache 2.0 coding model | Codex and OpenAI API | Claude Code and Anthropic API |
| Best viewed as | Enterprise RAG + sovereignty + open model control | Generalist + consumer | Reasoning + coding specialist |
Failure modes
- No consumer chat app. Cohere does not publish a user-facing assistant. Teams evaluating for solo use should look elsewhere.
- Open model does not remove deployment work. Command A+ is Apache 2.0, but teams still need inference infrastructure, evals, monitoring, and support.
- Coding model does not equal coding product. North Mini Code is useful for teams that own a harness, model serving, evals, and policy. It is not a packaged developer workflow like Cursor, Copilot, Claude Code, Codex, Devin, or Windsurf.
- Peak reasoning needs testing.-heavy work, but buyers should benchmark it against Claude Opus 4.8 and OpenAI frontier models on their own hardest tasks.
- Developer ecosystem is smaller. Fewer SDKs, fewer community integrations, and a narrower tutorial corpus than OpenAI or Anthropic.
- Pricing opacity on Command A+ and North Mini Code. Public pricing lists Command R and R+ rows more clearly than newer managed model usage. Confirm rates before projecting production cost.
- Rerank query math can surprise teams. Cohere defines one search as one query with up to 100 documents, but documents over 500 tokens are chunked and each chunk counts toward the ranked-document total.
- Model Vault pricing is per-instance, not per-token. $2,500-$3,250/mo per model per tier. Economical only for workloads that justify dedicated infrastructure.
- No image or video output. Workflows needing multimodal generation need a second tool alongside Cohere.
- Region and residency caveats. Availability of Model Vault in specific sovereign clouds depends on partner contracts; confirm your target region before signing. The reported Aleph Alpha deal could improve the Europe story, but buyers still need written deployment terms.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility x Value x Moat x Longevity, unweighted average). Last verified 2026-06-18 against Cohere pricing, Cohere models docs, Cohere release notes, North Mini Code, Command A+, the Cohere Transcribe page, and Cohere and Aleph Alpha’s announcement.
FAQ
Is Cohere free to use? Yes for prototyping. Trial API keys are free and rate-limited. Production API keys are pay-as-you-go, but new deployments should check current model rows rather than assuming legacy Command Light availability. Command R+ 08-2024 is listed at $2.50 input / $10 output per million tokens, and enterprise discounts bring R+ down to $1 input / $3 output per million tokens.
What is the current Cohere flagship?
Command A+ (command-a-plus-05-2026), released May 20, 2026. It is a 218B sparse MoE with 25B active parameters, 128K input context, 64K max generation, vision input, reasoning, tool use, 48-language support, and an Apache 2.0 license. Command R+ remains the production RAG workhorse at a lower price point.
What is North Mini Code? North Mini Code is Cohere’s first developer-facing coding model. It is a 30B total / 3B active MoE with Apache 2.0 weights, 256K total context, 64K max generation, and availability through Hugging Face, Cohere API, OpenRouter, and Model Vault. Treat it as model infrastructure for coding agents, not as a finished IDE assistant.
Does Cohere have a consumer chat app? No. Cohere sells to enterprise buyers and does not ship a consumer-facing assistant competitive with ChatGPT or Claude. North is an enterprise agent workspace, not a consumer chat product.
What is Model Vault? Cohere’s dedicated model inference platform, launched September 2025. Enterprises deploy Command, Rerank, and Embed inside isolated VPCs or on-prem infrastructure. Data never leaves the customer network. Pricing starts at $2,500/mo per model per tier.
How does Cohere compare to OpenAI for RAG? Cohere’s dedicated retrieval stack (Embed v4 + Rerank 4) outperforms OpenAI embeddings on public cross-lingual and long-document benchmarks. OpenAI wins on consumer chat, image generation, and plugin ecosystem. If retrieval accuracy is the bottleneck, Cohere is the stronger pick.
Sources
- Cohere pricing: Production API rates and Model Vault pricing
- Cohere release notes: Command A+, North Mini Code, Command R+, R, R7B, Embed, Rerank, and Transcribe updates
- Cohere models docs: current model IDs, context windows, modalities, and deprecation state
- Cohere Embed v4 page: Multimodal embedding model specifications
- Cohere Command A+ announcement: Apache 2.0 Command A+ release and model details
- Cohere North Mini Code announcement: Apache 2.0 coding-model release and availability
- Cohere Transcribe page: New 03-2026 enterprise speech-to-text model
- Cohere pricing docs: Token counting and billing mechanics
- Cohere and Aleph Alpha announcement: planned sovereign-AI combination
Related
- Category: AI Chatbots · AI Search
- Alternatives: Claude · ChatGPT · Gemini · Llama · Mistral AI · DeepSeek
