Cohere Command A+: Inside the 218B‑Parameter Open‑Source Model Built for Sovereign Enterprise AI

These capabilities are designed for enterprise use cases such as retrieval‑augmented generation (RAG), automated knowledge workflows, and tool‑using AI agents that interact with internal systems.

Architecture: 218B Parameters With Mixture‑of‑Experts Efficiency

The most distinctive aspect of Command A+ is its Mixture‑of‑Experts architecture.

• 218B total parameters across the model
• ~25B active parameters per token selected dynamically during inference

In MoE systems, only a subset of specialized neural “experts” is activated for each token processed. That design reduces compute requirements compared with dense models where all parameters are used for every token.

Cohere describes the approach as delivering high performance with minimal compute overhead, especially for enterprise workloads where cost and deployment flexibility matter as much as raw model size.

Deployment Requirements and Hardware Efficiency

Despite its large parameter count, Command A+ is designed to run on relatively limited hardware due to its sparse activation pattern.

Reported deployment configurations include:

• A single Nvidia B200 GPU or
• Two Nvidia H100 GPUs

For a model with over 200B parameters, this requirement is unusually modest and reflects the efficiency benefits of MoE inference.

Cohere also provides the model through its Model Vault and Chat API infrastructure for production deployments.

The practical implication is that organizations can run the model inside their own infrastructure, including private clouds or on‑premise environments, instead of sending sensitive data to external AI APIs.

Licensing and Open‑Source Availability

Command A+ is released under the Apache 2.0 license, one of the most permissive open‑source licenses used in software and AI.

This means:

• Commercial use is permitted
• Organizations can modify and deploy the model
• No additional model‑specific usage restrictions are imposed

Many recent “open‑weight” AI releases include restrictions on commercial use or scale, so the Apache 2.0 choice is a notable departure from that trend.

Performance Positioning

Cohere states that Command A+ surpasses previous models in the Command family and integrates capabilities that were previously distributed across multiple specialized models such as Command A Reasoning and Command A Vision.

Public release materials emphasize improvements in:

• Tool‑use and agentic tasks
• Multimodal reasoning
• Multilingual performance

However, publicly available sources describing the launch do not provide detailed benchmark tables or exact score comparisons, so independent quantitative comparisons with earlier models remain limited in current documentation.

Why Command A+ Matters for Sovereign AI

Cohere frames Command A+ as part of a larger push toward sovereign AI—systems that governments and enterprises can run under their own jurisdiction and infrastructure.

This concept emphasizes:

• Full control over training and inference environments
• Local data residency and regulatory compliance
• Reduced dependence on foreign cloud providers

Cohere explicitly markets the model as suitable for “sovereign critical infrastructure” deployments for governments and regulated sectors such as finance, healthcare, and energy.

The Broader Strategic Context

The release of Command A+ aligns with several strategic moves by Cohere aimed at building a sovereign AI ecosystem:

Transatlantic merger with Aleph Alpha
Cohere announced plans to merge with Germany‑based AI company Aleph Alpha to create a transatlantic AI provider focused on enterprise and public‑sector deployments.

Major investment from Schwarz Group
Companies within Germany’s Schwarz Group committed $600 million (€500M) in financing tied to Cohere’s upcoming funding round and the broader sovereign‑AI initiative.

Expansion into regulated industries
Cohere also acquired Reliant AI to expand sovereign enterprise AI offerings in sectors such as healthcare and biopharma.

Together, these moves indicate that Command A+ is not just a model release—it is a building block in a broader attempt to create an independent AI stack serving governments and enterprises outside traditional hyperscaler ecosystems.

The Bottom Line

Command A+ represents a combination of architectural efficiency and strategic positioning.

Technically, the model combines a large MoE architecture, multimodal capabilities, long context, and enterprise tooling while maintaining relatively modest deployment requirements. Strategically, it reinforces Cohere’s push to become a major provider of sovereign, privately deployable AI systems for governments and regulated industries worldwide.

If that strategy succeeds, Command A+ may be remembered less as a single model release and more as a key step in the emergence of alternative AI infrastructure outside the dominant cloud ecosystems.