Command A+: Making sovereign agentic capabilities available to all (cohere.com)

Cohere has launched Command A+, an open-source mixture-of-experts (MoE) model optimized for high-performance agentic tasks, available under the Apache 2.0 license. This highly versatile large language model (LLM) boasts 218 billion total parameters, with 25 billion active ones, and is designed for efficient deployment across a variety of enterprise applications, including multimodal processing and reasoning. The model can support inputs in 48 languages and is engineered for practical use, allowing developers to run it on minimal hardware with impressive efficiency and low compute overhead. It also features advanced tokenization that improves speed and reduces inference costs, making it more accessible for diverse operational environments. The significance of Command A+ lies in its push for "sovereign AI," which empowers developers by providing control over capable AI tools that can be run locally. Performance benchmarks indicate significant improvements over prior Command models, particularly in reasoning, multilingual tasks, and multimodal understanding. Notable enhancements include up to 63% higher output rates and reduced latency compared to previous versions. This advancement reflects Cohere’s commitment to fostering a self-sufficient AI ecosystem that meets the evolving demands of enterprise users, thus setting the stage for broader adoption of high-performance AI solutions in various industries.