Skip to content
Main Site News Console

Granite Embedding Multilingual R2: Open-Source Apache 2.0 Multilingual Embedding Model with 32K Context — Best Retrieval Quality Under 100M Parameters

· Hugging Face Translated
教程模型卡

Granite Embedding Multilingual R2: An Apache 2.0 Open-Source Multilingual Embedding Model with 32K Context Support — Best Retrieval Performance Under 100M Parameters

Enterprise article
Published on May 14, 2026

TL;DR: Two new multilingual embedding models built on ModernBERT and released under the Apache 2.0 license: one is a lightweight 97M-parameter model that outperforms all open-source multilingual embedding models under 100M parameters on the MTEB multilingual retrieval task (60.3); the other is a full-size 311M-parameter model that scores 65.2 on the MTEB multilingual retrieval task (ranking #2 among open-source models under 500M parameters) and supports Matryoshka. Both cover 200+ languages, have been tuned on 52 languages, support 32K token context — 64× that of R1 — and add code retrieval capability for 9 programming languages.

Image 16: Granite Embedding Multilingual R2

In this article: Enterprise-ready by design · A strong sub-100M multilingual model · What changed from R1 · Training the full-size 311M model · Building the compact 97M multilingual model