IBM Granite Embedding Multilingual R2: Apache 2.0 Multilingual Embeddings with 32K Context

Summary

IBM released Granite Embedding Multilingual R2, two Apache 2.0 licensed embedding models built on ModernBERT. The compact **97M-parameter model** (384-dim) scores 60.3 on MTEB Multilingual Retrieval — the best sub-100M multilingual embedder available, outperforming multilingual-e5-small (50.9) by 9.4 points at 3x smaller size. The full **311M model** (768-dim) scores 65.2 on MTEB Multilingual Retrieval (#2 among open models under 500M parameters) and supports Matryoshka dimension reduction down