IBM Granite

IBM Granite is an enterprise-grade open-source model family built with transparent training data provenance and IP indemnification — critical features for regulated industries where auditability and licensing clarity are non-negotiable.

Key use cases include:

• Enterprise document processing in regulated industries (finance, healthcare, legal)
• Code generation with Granite Code models trained on permissively-licensed code
• Time-series forecasting with specialized Granite Time Series models
• Applications requiring clear IP provenance and audit trails
• On-premise deployments in air-gapped environments
• RAG and knowledge management within the IBM ecosystem

Granite is used by enterprises — particularly in financial services, healthcare, and government — where the provenance of training data matters for compliance. IBM provides indemnification against IP claims, reducing legal risk for enterprise adoption.

Available through IBM watsonx.ai (hosted) and as open-source weights on Hugging Face under the Apache 2.0 license.

1. Access through IBM watsonx.ai by creating an IBM Cloud account, or download open-source weights from Hugging Face.
2. For watsonx.ai:

   pip install ibm-watsonx-ai
   from ibm_watsonx_ai.foundation_models import ModelInference
   model = ModelInference(
       model_id='ibm/granite-3-8b-instruct',
       credentials={'url': 'https://us-south.ml.cloud.ibm.com',
                    'apikey': 'your-ibm-cloud-api-key'},
       project_id='your-project-id'
   )

3. For local use:

   ollama run granite3.1:8b

Pricing: Open-source weights are free (Apache 2.0). watsonx.ai offers a free lite tier, with production plans starting at pay-as-you-go rates. Enterprise pricing with IP indemnification available. See IBM watsonx pricing.