Granite Embedding Multilingual R2 Released as Open Source

IBM and Hugging Face have jointly released Granite Embedding Multilingual R2, an open-source multilingual embeddings model that sets a new benchmark for retrieval and embedding tasks across languages. Available under the permissive Apache 2.0 license, the model is designed to be freely used, modified, and deployed by developers and enterprises worldwide. One of the standout features of Granite Embedding Multilingual R2 is its 32,000-token context window, which allows it to process and understand significantly longer documents than many competing models. This extended context is particularly valuable for applications such as retrieval-augmented generation (RAG), where the ability to encode entire documents or lengthy passages can dramatically improve the accuracy and relevance of search results. The model achieves best-in-class performance for multilingual retrieval and embedding tasks, meaning it can effectively handle content in dozens of languages with high fidelity. This makes it an ideal foundation for building cross-lingual AI applications, such as global search engines, multilingual customer support systems, and document analysis tools that need to work across language barriers. By releasing the model as open source, IBM and Hugging Face are democratizing access to state-of-the-art embedding technology. Developers can integrate Granite Embedding Multilingual R2 into their own pipelines without worrying about licensing costs or restrictions, fostering innovation in areas like semantic search, knowledge management, and AI-powered content discovery. For enterprises operating in multilingual environments, this release is particularly timely. As AI adoption accelerates globally, the need for models that can understand and retrieve information across languages has never been greater. Granite Embedding Multilingual R2 provides a powerful, free, and open solution that meets this need while maintaining high performance standards.