Introduction
Canadian AI lab Cohere has announced the release of Command A+, a groundbreaking sparse mixture-of-experts (MoE) model designed specifically for agentic tasks. With 218 billion total parameters and only 25 billion active at inference, Command A+ marks Cohere's first model released under the permissive Apache 2.0 license. This move amplifies the company’s push toward open, efficient AI systems that can autonomously reason, plan, and execute multi-step workflows.
What Is Command A+?
Command A+ is a large language model (LLM) built on a sparse MoE architecture. The model boasts a total of 218 billion parameters, but only 25 billion are activated per forward pass, making it both powerful and computationally efficient. This design allows the model to handle complex reasoning tasks while keeping inference costs and latency low.
Sparse MoE Architecture
Sparse mixture-of-experts divides the model into multiple specialized sub-networks (experts). During inference, only a subset of these experts is activated based on the input. This approach delivers the quality of a much larger dense model while using a fraction of the compute. Command A+ employs this technique to excel in agentic tasks, where the model must break down problems, call tools, and maintain context over long chains of actions.
Designed for Agentic Tasks
Cohere specifically optimized Command A+ for agentic AI use cases. Unlike standard chatbots that generate only text, agentic models are designed to:
- Plan and execute multi-step workflows
- Call external APIs and tools
- Reason about intermediate states and adapt
- Maintain long-term context across interactions
This makes Command A+ suitable for applications like autonomous code generation, data pipeline orchestration, and self-driving AI assistants that can complete tasks with minimal human intervention. The 25 billion active parameters ensure that the model remains responsive and cost-effective while handling these demanding scenarios.
First Apache 2.0 License for Cohere
Command A+ is released under the Apache 2.0 license, a permissive open-source license that allows anyone to use, modify, and distribute the model, even for commercial purposes, as long as they include attribution. This is Cohere’s first model under this license, signaling a major shift toward openness. Previously, Cohere’s models (like Command R) used a more restrictive license that limited commercial use or required special arrangements.
The Apache 2.0 choice has significant implications:
- Adoption acceleration – Developers can integrate Command A+ into their products without legal friction.
- Community contributions – Researchers can fine-tune and improve the model openly.
- Competitive positioning – Cohere directly challenges other open models like Meta’s Llama 3.1 or Mistral AI’s offerings, but with a focus on agentic tasks.
Context: Cohere's Recent Merger with Aleph Alpha
The release of Command A+ comes shortly after Cohere announced a merger with German AI startup Aleph Alpha. Aleph Alpha is known for its sovereign AI capabilities, particularly in Europe. The merger is expected to combine Cohere’s model architecture expertise with Aleph Alpha’s strength in enterprise-scale, privacy-preserving AI solutions.
This strategic move likely influenced the development of Command A+. By merging, Cohere gains access to Aleph Alpha’s research on sparse attention and efficient inference, which aligns perfectly with the MoE design of Command A+. Additionally, the combined entity can tap into Europe’s growing demand for open, transparent AI systems—a demand that Apache 2.0 licensing directly addresses.
Implications for the AI Landscape
Command A+ arrives at a pivotal moment in the AI industry. As organizations move beyond simple chat interfaces toward agentic workflows, models that can autonomously execute tasks become critical. Cohere’s choice of a sparse MoE architecture ensures that these advanced capabilities are not limited to cloud giants with massive budgets. Smaller companies and individual developers can now run a state-of-the-art agentic model on moderate hardware.
Moreover, the Apache 2.0 license lowers the barrier for innovation in verticals such as healthcare, finance, and logistics, where agentic AI can automate complex processes under strict compliance needs. The merger with Aleph Alpha further strengthens Cohere’s ability to deliver sovereign AI solutions that keep data within national borders—a key requirement for many governments.
Sovereign AI and Open-Source Synergy
By combining open licensing with the Aleph Alpha merger, Cohere positions itself as a leader in sovereign AI—AI systems that are controlled, auditable, and deployable within a specific jurisdiction. Command A+ under Apache 2.0 allows enterprises to inspect, modify, and host the model on their own infrastructure, reducing reliance on foreign cloud providers.
Conclusion
Cohere’s Command A+ represents a significant step forward in making powerful, agentic AI accessible to everyone. With its sparse MoE design, 218B total parameters (25B active), and the permissive Apache 2.0 license, the model is poised to accelerate the development of autonomous AI agents across industries. The simultaneous merger with Aleph Alpha underscores Cohere’s commitment to open, sovereign, and efficient AI. As developers and enterprises begin to experiment with Command A+, the era of agentic AI is officially open for business.