Model Release
Cohere Releases Command A+ Open-Source Model
Cohere launched Command A+, a 218B parameter MoE model, under Apache 2.0 license for enterprise agentic tasks.
Image: Cohere
Cohere today announced the release of Command A+, an open-source mixture-of-experts (MoE) model designed for high-performance agentic tasks with minimal compute overhead. The model, available under an Apache 2.0 license, unifies capabilities from previous Command generations into a single scalable model. Command A+ is optimized for reasoning, agentic workflows, RAG, multilingual, and multimodal document processing, supporting 48 languages. Developers can access the model on Hugging Face with several near lossless quantizations or deploy it in Model Vault for a managed inference environment. *Source: [cohere](https://cohere.com/blog/command-a-plus)*
Command A+ outperforms earlier Command A models in key enterprise workload dimensions, including multimodal understanding, retrieval, long-horizon reasoning, and complex reasoning. It achieved a score of 37 on the Artificial Analysis Intelligence Index, outperforming other leading open models. The model is also the most efficient to date, with 218B total parameters and 25B active parameters, compared to Command A Reasoning’s 111B dense architecture. At the same quantization and concurrency levels, Command A+ delivers up to 63% higher Output Tokens per Second (TOPS) and reduces Time To First token (TTFT) by up to 17%. *Source: [cohere](https://cohere.com/blog/command-a-plus)*
Cohere emphasized that Command A+ is engineered for practical, developer-focused use, including support for low-bit quantization, efficient inference, and integration across open inference frameworks. The model is available in 16-bit (BF16), 8-bit (FP8), and 4-bit (W4A4) quantizations, enabling it to run on as little as two NVIDIA H100s or a single NVIDIA Blackwell GPU with minimal quality degradation. *Source: [cohere](https://cohere.com/blog/command-a-plus)*
Key points
- Cohere released Command A+, a 218B parameter mixture-of-experts (MoE) model, under an Apache 2.0 license.
- Command A+ outperforms previous Command A models in key enterprise workload dimensions, including multimodal understanding, retrieval, long-horizon, and complex reasoning.
- Command A+ achieved a score of 37 on the Artificial Analysis Intelligence Index, outperforming other leading open models.
- The model is available in 16-bit (BF16), 8-bit (FP8), and 4-bit (W4A4) quantizations, enabling it to run on as little as two NVIDIA H100s or a single NVIDIA Blackwell GPU.
- Command A+ delivers up to 63% higher Output Tokens per Second (TOPS) and reduces Time To First token (TTFT) by up to 17% compared to Command A Reasoning.
- Cohere emphasized that Command A+ is optimized for practical, developer-focused use, including support for low-bit quantization and efficient inference.