Granite Embedding
IBMGranite Embedding is IBM's retrieval model line for enterprise RAG, currently in its R2 generation. The R2 release ships two English bi-encoders on a ModernBERT backbone (47M and 149M parameters, 8192-token context), two multilingual bi-encoders (97M and 311M) and an English cross-encoder reranker. IBM's pitch is provenance as much as accuracy: the models are trained on permissively licensed open datasets plus IBM-collected data, and evaluated on enterprise-shaped retrieval - tables, long documents, conversational multi-turn and code - as well as BEIR and MTEB.
Openness
3 high confidence- weights
- open(four ungated Apache-2.0 R2 embedding checkpoints, 47M to 311M)
- data
- closed(permissively licensed public datasets plus select proprietary IBM data, neither enumerated nor shipped)
- code
- partial(usage and documentation in ibm-granite/granite-embedding-models
- license
- Apache-2.0(OSI)
Every R2 checkpoint ships under Apache-2.0, ungated. The training mixture -- permissively licensed public sources plus proprietary IBM data -- is described only by category and ships no dataset or script, and the public repository holds documentation and usage examples rather than a training pipeline.
- https://huggingface.co/api/models/ibm-granite/granite-embedding-english-r2 recorded 2026-09-11
gated: false; private: false; cardData license apache-2.0.
- https://huggingface.co/ibm-granite/granite-embedding-english-r2/raw/main/README.md recorded 2026-09-11
license: apache-2.0 in the front matter and named again under Model Details; training data described as permissively licensed open-source plus select proprietary data, with no dataset listing; the repository is given as ibm-granite/granite-embedding-models.
- https://raw.githubusercontent.com/ibm-granite/granite-embedding-models/main/README.md recorded 2026-09-11
The repository introduces the collection, links the Hugging Face models and IBM documentation, and states the Apache-2.0 offer; it carries no pretraining or contrastive-training pipeline.
Adoption
4 high confidenceAdoption is measured as Hugging Face downloads across the granite-embedding R2 checkpoints, led by the small English model. The R2 reranker checkpoint is excluded because every other vendor in this category has its reranker scored as a separate product, and the earlier R1 line is excluded as a prior generation.
- https://huggingface.co/api/models/ibm-granite/granite-embedding-311m-multilingual-r2 recorded 2026-09-11
downloads: 117691; gated: false; private: false.
- https://huggingface.co/api/models/ibm-granite/granite-embedding-97m-multilingual-r2 recorded 2026-09-11
downloads: 99229; gated: false; private: false.
- https://huggingface.co/api/models/ibm-granite/granite-embedding-english-r2 recorded 2026-09-11
downloads: 52733; gated: false; private: false.
- https://huggingface.co/api/models/ibm-granite/granite-embedding-small-english-r2 recorded 2026-09-11
downloads: 6324429; gated: false; private: false.
Capability
3 high confidenceGranite Embedding R2 leads its size class on the enterprise-shaped tasks IBM built it for -- table retrieval, long documents and conversational multi-turn -- but its scope stays narrow: the strongest checkpoint is 149M parameters and English-only, scoring 62.8 on a 41-task benchmark, and the multilingual checkpoints are smaller still.
- https://huggingface.co/ibm-granite/granite-embedding-english-r2/raw/main/README.md recorded 2026-09-11
Evaluation table reports granite-embedding-english-r2 at 53.1 BEIR retrieval (15), 62.8 MTEB-v2 (41), 55.3 CoIR and 56.7 MTRAG, with the small-english-r2 variant at 50.9 / 61.1, and a second table showing it ahead of gte-modernbert-base and snowflake-arctic-embed-m-v2.0 on the enterprise-shaped average.
Verified 2026-09-11