Sovereign enterprise AI
Run a capable agentic model inside infrastructure controlled by the organization or a dedicated private environment.
Independent tool overview
Command A+ at a glance
Command A+ is Cohere's open-source, mixture-of-experts model for enterprise reasoning, agents, retrieval, image understanding, and multilingual work across 48 languages.
Visit the official Command A+ site ↗
- Model ID
- command-a-plus-05-2026
- Architecture
- 218B total / 25B active sparse mixture of experts
- License
- Apache 2.0
- Context window
- 128,000 tokens
- Maximum output
- 64,000 tokens
- Inputs and output
- Text and images in; text out
- Languages
- 48
- Knowledge cutoff
- April 1, 2025
- Minimum listed hardware
- 1 B200 or 2 H100 GPUs with W4A4 quantization
- Last reviewed
- August 29, 2026
Overview
What Command A+ is
Command A+ is the final and most capable model in Cohere's Command A family. It combines capabilities that were previously split across Command A, Command A Reasoning, Command A Vision, and Command A Translate into one set of weights.
The model accepts text and image inputs and produces text, reasoning, citations, structured output, and tool calls. It is designed for enterprise agents, retrieval-augmented generation, document and chart analysis, multilingual workflows, and private deployments where organizations need to control the model and data path.
Cohere releases the weights under Apache 2.0 and also provides hosted API access and private Model Vault deployment. Command A+ is a large 218-billion-parameter sparse mixture-of-experts model, with 25 billion active parameters per token, so self-hosting still requires data-center-class GPUs.
Use cases
Who Command A+ is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Multilingual agents
Build tool-using workflows across 48 languages, including all official European Union languages and major Asian and Middle Eastern languages.
Document and chart reasoning
Analyze text and image-based business materials, then return grounded answers, citations, or structured data.
Retrieval-augmented generation
Use Cohere's enterprise-oriented instruction following, citations, and tool support to answer from controlled knowledge sources.
Private coding and operations agents
Deploy an open model for multi-step tool use where source data or application access cannot move into a shared public service.
Capabilities
Core Command A+ features
Unified Command capabilities
Combines reasoning, vision input, tool use, retrieval, multilingual understanding, and translation within one model rather than routing among separate Command variants.
Multimodal reasoning
Accepts images alongside text for documents, charts, diagrams, screenshots, and other visual enterprise inputs.
Agentic tool use
Supports tool calls and long-horizon workflows where the model must choose actions, interpret results, and continue toward a goal.
Citations and structured output
Can ground responses in supplied sources and return machine-readable schemas for retrieval and business automation pipelines.
Controllable reasoning
Supports thinking modes and token budgets so developers can balance reasoning depth, latency, and compute.
Open deployment
Apache-2.0 weights are available in BF16, FP8, and W4A4 variants for vLLM and Transformers-based serving.
Efficient MoE inference
Activates 25 billion of 218 billion parameters per token and can run in its lowest listed configuration on one B200 or two H100 GPUs.
Process
How the Command A+ workflow works
Step 1
Choose hosted or self-managed deployment
Use the Cohere API for evaluation, Model Vault for dedicated managed inference, or the Apache-2.0 weights when infrastructure and data must remain under your control.
Step 2
Select the right quantization
Match BF16, FP8, or W4A4 to available accelerators, accuracy requirements, throughput targets, and operational budget.
Step 3
Define grounding and tools
Provide retrieval sources, citations, function schemas, and explicit approval boundaries instead of asking the model to improvise business actions.
Step 4
Test every required language and modality
Build a representative evaluation set for documents, screenshots, retrieval, tool calls, translations, and edge cases in each production language.
Step 5
Tune reasoning and output controls
Set thinking budgets, structured schemas, safety mode, and maximum output length based on the task's risk and latency needs.
Step 6
Monitor production behavior
Track citations, tool-call validity, task success, latency, GPU utilization, and cost; keep human approval around financial, legal, security, and external-write actions.
Cost
Command A+ pricing and free plan
Cohere's current Command A+ model page says API use is free for trial and production keys until rate limits are reached. Dedicated production deployment is sold through Model Vault, while self-hosting has no model license fee but requires substantial GPU and operating costs.
Cohere API
Free within rate limits
Hosted evaluation and limited production access using command-a-plus-05-2026.
- Trial and production keys supported
- Rate limits apply
- Chat API endpoints
- Confirm current quota before capacity planning
Model Vault
Custom
Dedicated managed deployment for enterprise production workloads.
- Private managed inference
- Production capacity and terms by agreement
- Enterprise deployment support
- Contact Cohere for pricing
Self-hosted weights
No model license fee
Apache-2.0 model weights operated on your infrastructure.
- W4A4 minimum: 1 B200 or 2 H100 GPUs
- Hardware, cloud, power, storage, and engineering billed separately
- vLLM and Transformers support
- Organization owns serving and security operations
Pricing checked . Check current pricing at the source ↗
Assessment
Command A+ strengths and limitations
Where it stands out
- One open model covers reasoning, image understanding, tool use, retrieval, and multilingual work
- Apache 2.0 licensing supports private modification and commercial deployment
- Forty-eight-language coverage is unusually broad for an enterprise agentic model
- Large 128K input and 64K output limits support substantial documents and long-form structured work
- Sparse MoE design lowers active compute relative to its 218B total parameter count
- Hosted, dedicated, and self-managed deployment paths support different sovereignty requirements
What to consider
- Self-hosting still requires expensive data-center GPUs and experienced inference operations
- The model accepts images but not native audio or video input and produces text rather than generated media
- Its April 2025 knowledge cutoff requires retrieval or tools for current information
- Free API access is rate-limited and is not a substitute for a production capacity agreement
- Open weights do not remove the need for security testing, safety controls, monitoring, and legal review
- Tool calls, citations, translations, and structured outputs must be evaluated on the organization's real languages and workflows
Compare
Command A+ alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Consumer
Gemini 3.1 Pro
Choose Gemini 3.1 Pro for Google's hosted multimodal reasoning, one-million-token context, and broad search and tool integrations.
Explore Gemini 3.1 Pro →Consumer
Claude Opus 4.8
Choose Claude Opus 4.8 for Anthropic's current top hosted model when frontier coding and agent reliability matter more than open self-deployment.
Explore Claude Opus 4.8 →Business Operations
DeepSeek
Choose DeepSeek for another open-model ecosystem with accessible reasoning options and broad third-party deployment support.
Explore DeepSeek →Questions
Command A+ FAQs
What is Command A+?
Command A+ is Cohere's 218B-total, 25B-active mixture-of-experts model for reasoning, vision input, agents, retrieval, structured output, and multilingual work across 48 languages.
Is Command A+ open source?
Yes. Cohere publishes the model weights under the Apache 2.0 license on Hugging Face, with BF16, FP8, and W4A4 variants.
How much does Command A+ cost?
Cohere currently says its hosted API is free for trial and production keys until rate limits are reached. Model Vault uses custom enterprise pricing. Self-hosting has no model license fee but incurs GPU, infrastructure, and operating costs.
What hardware does Command A+ need?
Cohere lists a minimum of one B200 or two H100 GPUs for the W4A4 quantized model. Higher-precision variants require more hardware; the Hugging Face model cards list the current requirements for each release.
Can Command A+ analyze images?
Yes. It accepts text and image inputs and can reason over documents, charts, diagrams, and other visuals. Its output is text, reasoning, citations, structured data, or tool calls.
How is Command A+ different from Command A?
Command A+ is the later unified model. It adds reasoning and vision, expands language coverage from 23 to 48, improves agentic performance, and uses a sparse mixture-of-experts architecture.
Is Command A+ suitable for private deployment?
Yes. Sovereign deployment is a primary use case. Organizations can self-host the Apache-2.0 weights or use Cohere Model Vault for dedicated managed inference, subject to their security and commercial requirements.
Bottom line
Our Command A+ verdict
Command A+ is a strong candidate for organizations that need one privately deployable model to handle agents, retrieval, images, reasoning, and multilingual work. Its Apache-2.0 release and modest active-parameter count are meaningful advantages, but 'efficient' is relative: the minimum supported hardware is still enterprise-grade. Evaluate the hosted API first, then compare Model Vault with self-hosting using real throughput, language, tool, and compliance requirements.
Visit Command A+ website ↗Similar Business Operations Tools
View all Business Operations AI tools →
Claude for Small Business
Claude for Small Business - Anthropic's new connectors and workflows for running payroll, invoicing, and campaigns
Meta Business Agent
Meta Enterprise Agent - AI agent for customer sales and support across Meta's apps
Cofounder 2
Cofounder 2 - General Intelligence's agent orchestrator that manages full business operations across engineering, sales, marketing, and design
Mistral OCR 4.1
Mistral OCR 4 - Mistral’s OCR model with layout-aware document understanding
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.