Companies don't want to send their data to OpenAI. Full stop. They want to build RAG applications, but mentioning a model someone else hosts ends every conversation. This is the reality of building real-world applications. @Cohere just released a multilingual model optimized for RAG applications that's lightweight and fast (very fast!), and it looks like it finally cracked this problem! Command R7B is a small 7B-parameter open-weight model, and you can run it anywhere: • It runs on a low-end GPU • It runs on a MacBook • It even runs on a CPU The model supports tool integration and can tackle advanced reasoning, summarization, and question-answering use cases. And it's specialized in RAG use cases! This is a big deal! Multiple people have asked me about working on non-English use cases. While many models kinda work with several languages, Command R7B is the only one I know that's natively multilingual. Cohere trained it in 23 languages and extensively evaluated it in 10. Look at the attached benchmark. Command R7B comes on top across the board. It gives best-in-class performance in RAG and tool-use tasks.