Cohere Command A+ Now Available in Microsoft Foundry

Command A+ is Cohere’s newest mixture-of-experts (MoE) model designed for enterprise-grade agentic AI workloads with exceptional efficiency and scalability. Released under an Apache 2.0 license, the model brings together reasoning, multimodal understanding, multilingual capabilities, retrieval, and tool use into a single unified model optimized for real-world enterprise AI applications.

Built for Enterprise Agentic AI

Command A+ is optimized for practical enterprise workloads that require strong reasoning, orchestration, and operational efficiency.

The model is designed for scenarios including:

• Agentic workflows and autonomous systems
• Retrieval-Augmented Generation (RAG)
• Multimodal document understanding
• Enterprise search and knowledge retrieval
• Coding and business automation
• Multilingual enterprise applications

Command A+ supports:

• 128K context window
• Text, image, and tool-use inputs
• Reasoning and tool-use outputs
• Support for 48 languages
• vLLM and Transformers frameworks

Deploy with Managed Compute in Microsoft Foundry

With Managed Compute in Microsoft Foundry, organizations can deploy Command A+ on dedicated managed infrastructure while benefiting from a unified AI platform experience across development, deployment, evaluation, and operations.

Managed Compute enables customers to:

• Deploy open-source models on managed infrastructure without managing complex AI serving stacks
• Scale AI workloads with greater control over performance and infrastructure
• Optimize throughput, latency, and operational efficiency
• Standardize deployment and operations across models and workloads
• Integrate AI governance, monitoring, and evaluation workflows into production systems

This deployment model is particularly valuable for organizations looking to operationalize open-source models while maintaining enterprise-grade reliability and operational simplicity.

High Performance with Exceptional Efficiency

Efficiency was a core design goal for Command A+. The model uses a sparse MoE architecture with 218B total parameters and only 25B active parameters, allowing it to deliver strong enterprise performance while minimizing compute overhead.

According to Cohere, Command A+ can run on:

• A single NVIDIA Blackwell GPU
• Or as little as two NVIDIA H100 GPUs using optimized quantization techniques

The model is also Cohere’s fastest model to date, delivering:

• Up to 63% higher output tokens per second
• Reduced latency and time-to-first-token
• Additional inference acceleration through speculative decoding
• Improved tokenizer efficiency across multilingual workloads

These optimizations help enterprises reduce infrastructure costs while scaling production AI systems more efficiently.

Strong Enterprise and Multimodal Capabilities

Command A+ consolidates capabilities that were previously spread across multiple Command model variants into a single model. The model demonstrates strong performance across:

• Long-horizon reasoning
• Agentic coding
• Multimodal understanding
• Tool use
• Enterprise retrieval tasks
• Multilingual reasoning and translation

Cohere reports significant improvements across enterprise-focused benchmarks and internal agentic evaluations, including gains in:

• Agentic question answering
• Spreadsheet analysis
• Memory-based reasoning
• Document understanding
• Multilingual reasoning

Open, Flexible, and Enterprise Ready

Command A+ is available under an Apache 2.0 license, giving organizations flexibility to customize, adapt, and deploy the model for enterprise workloads.

With availability through Microsoft Foundry Managed Compute, enterprises can combine the flexibility of open-source AI with the operational benefits of managed deployment infrastructure and a unified AI platform experience.

Get Started

Command A+ is now available in Microsoft Foundry as a Managed Compute offer. Try it today AI​.azure.com