FrontierMay 21, 2026via MarkTechPost
Cohere Releases Command A+: A 218B Sparse MoE Model for Agentic Workflows That Runs on as Few as Two H100 GPUs
Why it matters
Cohere's Command A+ consolidates four models into one sparse MoE that dramatically reduces compute requirements for agentic workflows, challenging the inference cost economics that have favored smaller closed models. This is a direct shot at deployment efficiency—the real bottleneck for enterprises scaling AI agents.
Key signals
- 218B parameter Sparse Mixture-of-Experts model
- Runs on as few as two H100 GPUs at W4A4 quantization
- Consolidates four prior Command A variants into single model
- First multimodal reasoning model from Cohere
- Supports 48 languages
- Open-source release
- Designed for agentic workflows
The hook
218B parameters. Two H100s. Cohere just made enterprise agentic AI radically cheaper to deploy.
Cohere releases Command A+, an open-source 218B Sparse Mixture-of-Experts model consolidating four prior Command A variants into one. It runs on as few as two H100 GPUs at W4A4 quantization, supports 48 languages, and is Cohere's first multimodal reasoning model.