Granite Embedding Multilingual R2 Models

TL;DR

The paper presents the Granite Embedding R2 models, advanced multilingual encoder-based embeddings for dense retrieval across 200+ languages, with state-of-the-art performance and enterprise-focused design.

Contribution

Introduction of two multilingual encoder-based embedding models with expanded language support, larger context window, and improved retrieval performance, including a compact high-performing variant.

Findings

Achieved state-of-the-art performance in multilingual and cross-lingual retrieval tasks.

Developed a compact model with the highest retrieval score under 100M parameters.

Supported enterprise use with governance and open licensing.

Abstract

We introduce the multilingual Granite Embedding R2 models, a family of encoder-based embedding models for enterprise-scale dense retrieval across 200+ languages. Extending our English-focused R2 release, these models add enhanced support for 52 languages and programming code, a 32,768-token context window (a 64x expansion over R1), and state-of-the-art overall performance across multilingual and cross-lingual text search, code retrieval, long-document search, and reasoning retrieval datasets. The release consists of two bi-encoder models based on the ModernBERT architecture with an expanded multilingual vocabulary: a 311M-parameter full-size, and a 97M-parameter compact model built via model pruning and vocabulary selection that achieves the highest retrieval score of any open multilingual embedding model under 100M parameters. The full-size also supports Matryoshka Representation Learning for flexible embedding dimensionality. Both models are trained on enterprise-appropriate data with governance oversight, and released under the Apache 2.0 license at https://huggingface.co/collections/ibm-granite, designed to support responsible use and enable unrestricted research and enterprise adoption.