What Is Cohere Command A+? The Sovereign Enterprise AI Model

Cohere's First Fully Open Frontier Model

Command A+ is Cohere's flagship open-weight AI model, released May 20, 2026 under the Apache 2.0 license. It's a decoder-only sparse mixture-of-experts model with 218 billion total parameters and about 25 billion active parameters, and it unifies four earlier Cohere models — Command A, Command A Reasoning, Command A Vision, and Command A Translate — into one scalable system covering reasoning, multilingual work, and multimodal document processing.

The headline feature is efficiency: Command A+ uses W4A4 quantization, which Cohere describes as lossless, to run on as few as two NVIDIA H100 GPUs or a single Blackwell B200. That's a small hardware footprint for a 218B-parameter model, and it's the whole point — Cohere built Command A+ so enterprises can deploy it inside a VPC, entirely on-premises, or fully air-gapped, wherever their sensitive data has to stay.

Built for Governments and Regulated Industries

Cohere frames this as a response to a specific worry: that open-weight AI development is increasingly concentrated in a small number of jurisdictions, namely China, leaving Western governments and regulated enterprises with few sovereign alternatives. A CISPE survey found roughly 72% of European enterprise IT decision-makers cite data sovereignty as a primary or secondary factor in cloud vendor selection — though analysts note that argument is strongest in specific verticals like financial services, healthcare, and the public sector, rather than universally decisive.

Command A+ vs Command A

Command A+ replaces four separate Cohere models with one unified system for enterprise deployments.

Independent latency measurements put Command A+ in the lowest-latency tier of open mixture-of-experts models in its weight class, alongside smaller Qwen 3.5 variants and NVIDIA's Nemotron Nano — a direct consequence of its 25B active parameters combined with W4A4 quantization. For agentic workloads that issue many short tool calls, that tokens-per-second advantage tends to matter more in practice than peak benchmark scores. Cohere's own MMLU, HellaSwag, and TruthfulQA figures were not public at launch, so those specific numbers are still worth checking before finalizing a deployment decision.

Why "Sovereign AI" Has Become a Real Procurement Category

Data sovereignty concerns have moved from a niche compliance footnote to a genuine driver of AI procurement decisions for governments and heavily regulated industries over the past couple of years, and Command A+'s positioning is a direct response to that shift. The core concern is straightforward: an API call to a foreign-hosted AI model means sensitive data — citizen records, financial transactions, defense-adjacent information — briefly or persistently passes through infrastructure outside a country's own legal jurisdiction, creating a compliance and national-security exposure that's independent of how well-secured that infrastructure actually is. For a growing number of government contracts, that exposure alone disqualifies foreign-hosted models regardless of capability.

Cohere's decision to release Command A+ fully open, rather than API-only, directly targets this specific buyer. A government agency or regulated bank can download the weights and run the model entirely within infrastructure they control, in the country they're legally required to keep the data in, without any data ever touching Cohere's own servers post-deployment. That's a materially different value proposition than a faster or cheaper API, and it explains why Cohere has built its enterprise business substantially around this positioning rather than competing head-on with OpenAI or Anthropic purely on general chatbot capability.

This trend isn't unique to Cohere — it's part of a broader "sovereign AI" movement that includes national AI infrastructure investments in countries like France, the UAE, and India, all motivated by similar concerns about strategic dependence on foreign AI providers. Command A+ is one of the more mature commercial products built specifically to serve that emerging market, rather than a general-purpose model that happens to also be open-weight.

For buyers evaluating whether this positioning actually applies to them, the honest test is whether data residency or air-gapped deployment is a genuine hard requirement, not a nice-to-have. If it is, Command A+ addresses a real constraint that closed frontier APIs simply can't solve regardless of price or capability. If it isn't — if the workload has no specific sovereignty requirement — the sovereignty angle is largely irrelevant, and the decision should come down to the same capability and cost trade-offs that apply to any other model choice.

Who Command A+ Is For

Frequently Asked Questions