Cohere Open Sources Command A+: 218B Parameter MoE Megamodel, Focusing on Enterprise Agents and Data Sovereignty

According to Dynasight Beating monitoring, Cohere has officially open-sourced the 218 billion-parameter sparse hybrid expert model Command A+, adopting the Apache 2.0 license to support enterprise-grade Agent development and private deployment. The new model is specifically designed to meet the data sovereignty requirements of critical infrastructure, enabling enterprises to deploy it with physical isolation without vendor lock-in.

The total parameter count of the new model is 218 billion (218B), but only 250 billion (25B) parameters are activated during a single inference, achieving a balance between computational efficiency and generative performance. The optimized inference efficiency allows it to run on just two NVIDIA H100 GPUs or a single B200 GPU, and Hugging Face has also simultaneously launched low-precision quantized versions such as W4A4, further reducing the deployment threshold.

Command A+ natively supports multimodal text-to-image input, providing a 128,000 (128K) Token input context window and an output length of 64,000 tokens. It has been deeply optimized for complex logical reasoning, autonomous tool invocation, database queries, and Agentic workflows, as well as long-document processing. The model supports 48 languages, including all official languages of the European Union.