On May 20, 2026, Cohere announced Command A+, a new Apache 2.0 Mixture-of-Experts model aimed at governments and regulated enterprises that want multimodal, multilingual, agent-capable AI without handing control to a closed vendor stack. The release combines vision inputs, reasoning, translation, and agentic tasks in one model, supports 48 languages, and is available through Cohere’s standard API endpoints with private deployment options for enterprise customers.
Cohere is framing the launch around sovereign critical infrastructure, not just model quality. That makes this a more commercially important release than a normal open-model update, because it targets the part of the market where data residency, deployment control, and compliance often matter more than raw benchmark theater.
What Cohere launched on May 20
Command A+ is Cohere’s newest Command-family model and its first Mixture-of-Experts release in that line. Cohere says the model activates 25 billion parameters out of 218 billion total parameters, offers a 128K input window with 64K output, and is now available under the model name command-a-plus-05-2026.
The product story is unusually compressed into one release. Instead of splitting vision, reasoning, translation, and agent workflows across separate models, Cohere is positioning Command A+ as a single enterprise model that can handle charts, PDFs, slides, financial documents, tool use, and multilingual work inside one governed deployment path. Cohere also says it is the strongest agentic model in the Command family.
- 48-language support, including all official EU languages
- Apache 2.0 licensing, giving buyers a clearer open-source path than many enterprise AI launches offer
- Low deployment footprint, with Cohere saying the model can run on as few as one B200 or two H100 GPUs
- Standard API availability now, plus private deployment options for enterprise customers
Why the sovereignty angle matters more than another spec sheet
The bigger signal is not that Cohere shipped another large model. It is that the company is explicitly tying the release to sovereignty at a moment when enterprise and public-sector buyers are asking harder questions about where models run, who controls updates, and what happens when critical workflows depend on someone else’s hosted stack.
In Cohere’s telling, Command A+ is built for organizations that need full visibility into model behavior, on-premises or private-cloud deployment, and tighter control over external data transmission. That framing puts the model in a different buying conversation from consumer-facing frontier launches. It is less about replacing a chatbot and more about making AI admissible inside regulated operations.
That matters for enterprises building AI agents because agent systems intensify governance concerns. A model that only answers questions is one thing. A model that can reason across internal documents, call tools, generate SQL, interpret charts, and operate across multilingual workflows is much closer to an execution layer. Once AI starts touching critical systems, sovereignty stops sounding like a policy slogan and starts sounding like an architecture requirement.
Business impact lands first in regulated agent deployments
The most immediate business impact should show up in the kinds of workloads Cohere is calling out directly: complex retrieval-augmented generation pipelines, multi-step SQL generation, financial document analysis, and multimodal enterprise document work. Those are not casual chat use cases. They are the kinds of jobs where buyers care about controllability, predictable infrastructure needs, and whether sensitive data can stay inside a private environment.
Document-heavy workflows get a more unified model option
Command A+ tries to reduce model sprawl for teams that have been stitching together separate systems for text reasoning, multilingual support, image understanding, and tool-oriented automation. If that works in practice, it could simplify how enterprises design internal research agents, audit workflows, reporting systems, and knowledge tools that need to interpret both text and visual business artifacts.
Multilingual compliance work becomes a stronger selling point
The 48-language coverage is a bigger detail than it first appears. For global businesses and public-sector teams, multilingual support is not only a user-experience feature; it affects rollout viability across regions, local operations, and regulated reporting environments. Cohere is clearly trying to make Europe and other multilingual markets part of the launch story, not an afterthought.
Compute efficiency becomes a deployment argument
Cohere’s claim that Command A+ can run on one B200 or two H100s is strategically important. Enterprise buyers evaluating private deployments often fail less on model quality than on hardware economics, latency targets, and the operational cost of keeping a model inside governed infrastructure. The more credible that efficiency story becomes, the more open models can move from lab experiments into real production AI agent stacks.
What to watch next
The next question is whether Command A+ earns adoption beyond Cohere’s sovereignty narrative. Buyers will want proof on real enterprise evaluations, especially for tool use, multimodal document work, and long-running agent tasks where operational reliability matters more than headline architecture details.
They will also want to see whether the open-source licensing and low-compute positioning translate into faster production rollout for regulated teams, or whether deployment still requires too much specialized infrastructure and model operations overhead. That adoption gap is where many promising enterprise AI releases still break down.
For AI agents, automation, and enterprise AI more broadly, the practical takeaway is clear: the market is shifting from a pure frontier-model race toward a control-and-deployment race. Cohere’s Command A+ matters because it makes open, governable, multilingual agent infrastructure a concrete product claim on May 20, 2026, not just a strategic talking point.