Granite Embedding Multilingual R2: Open Apache 2.0 Multilingual Embeddings with 32K Context — Best Sub-100M Retrieval Quality
IBM just open-sourced a sub-100M embedding model that matches enterprise retrieval quality. Here's why that matters for your RAG stack.
Why it matters
IBM's Granite Embedding Multilingual R2 demonstrates that smaller, open-source models can compete with larger proprietary alternatives on retrieval tasks, directly impacting enterprise AI infrastructure costs and vendor lock-in decisions.
The key facts
11 to knowModel size: Sub-100M parameters
Context window: 32K tokens
License: Apache 2.0 (open source)
Multilingual support
Positioning: Best-in-class retrieval quality for the sub-100M category
Use case: RAG (Retrieval-Augmented Generation) applications
Model: Granite Embedding Multilingual R2
Parameter size: Sub-100M
License: Apache 2.0 (open-source)
Positioning: Best-in-class retrieval quality for the parameter class
Published: May 14, 2026
Go to the source
Hugging Face Bloghuggingface.co