Introducing Command A+: Making sovereign agentic capabilities available to all - Cohere

2026-05-20 · Source: cohere.com via Google News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Advanced, medium

Summary

Cohere introduced Command A+, an open-source, mixture-of-experts (MoE) large language model released on May 20, 2026, under an Apache 2.0 license. Designed for complex reasoning, multimodal, and multilingual agentic tasks, Command A+ is Cohere's fastest and most powerful model, capable of running on as little as two NVIDIA H100 GPUs or a single B200 @ W4A4. It significantly outperforms previous Command series models, with τ²-Bench Telecom scores improving from 37% to 85% and Terminal-Bench Hard agentic coding performance reaching 25%. The model supports 48 languages and offers substantial gains in multimodal understanding, achieving 63% on MMMU Pro. Efficiency is enhanced by 4-bit quantization, delivering up to 63% higher Output Tokens per Second and reducing Time To First token by 17%, alongside a new tokenizer improving efficiency for non-European languages. Weights are available on Hugging Face in 16-bit, 8-bit, and 4-bit quantizations.

Key takeaway

For MLOps Engineers deploying enterprise-grade agentic AI, Command A+ offers a compelling open-source option. You should evaluate its 4-bit quantized version for efficient local deployment on minimal hardware like two H100s. This model's enhanced reasoning, multimodal, and multilingual capabilities, combined with significant speed and cost reductions, can accelerate your agentic workflow development and reduce inference expenses, especially for global applications. Consider integrating it via Hugging Face or Model Vault.

Key insights

Cohere's Command A+ democratizes enterprise-grade agentic AI through an efficient, open-source MoE model.

Principles

• Sovereign AI requires models that can be run, controlled, and adapted locally.
• Efficiency is a core constraint for practical enterprise AI deployment at scale.
• Advancing sovereign AI depends on simultaneous progress in performance, security, and cost.

Method

Command A+ utilizes a mixture-of-experts architecture, low-bit quantization (16-bit, 8-bit, 4-bit), speculative decoding, and a new tokenizer for efficient, high-performance inference.

In practice

• Deploy Command A+ on two H100s or one B200 GPU using 4-bit quantization.
• Integrate with vLLM or Transformers for agentic workflows.
• Utilize the new tokenizer for cost-effective multilingual inference.

Topics

• Command A+
• Agentic AI
• Mixture-of-Experts
• Model Quantization
• Multimodal LLM
• Multilingual AI
• Efficient Inference

Best for: AI Architect, NLP Engineer, Computer Vision Engineer, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• Cohere open-sources its strongest model yet — Cohere's Command A+ is a powerful, open-source MoE model designed for enterprise AI, offering strong performance and broad language support.
• Introducing Cohere Command A+ in Foundry — Cohere Command A+ in Microsoft Foundry unifies diverse AI capabilities for complex enterprise agentic workflows.
• Running local models is good now — Local LLMs have matured significantly, now enabling effective agentic coding and deep model introspection.
• Cohere Command A+ Now Available in Microsoft Foundry — Command A+ offers enterprise-grade agentic AI with high efficiency and managed deployment flexibility via Microsoft Foundry.
• 🗞️ Anthropic releases Claude Opus 4.8 on the same day as its $965B valuation round. — The AI frontier is rapidly expanding with new model capabilities, inference optimizations, and valuation milestones.
• North Mini Code: Agentic Coding Model for Developers - Cohere — Cohere's North Mini Code is an efficient, open-source agentic coding model designed for sovereign developer control.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cohere.com via Google News.