Granite 4.0 models

Granite 4.0 models are finetuned from their base models using a combination of open source instruction datasets with permissive license and internally collected synthetic datasets. They feature improved instruction following (IF) and tool-calling capabilities, making them more effective in enterprise applications.

Please Note: our micro model is an alternative option for users when mamba-2 support is not yet optimized.

Parameter Sizes

micro (3B)

ollama run ibm/granite4:micro

micro (3B)

ollama run ibm/granite4:micro-h

tiny (7B)

ollama run ibm/granite4:tiny-h

small (32B)

ollama run ibm/granite4:small-h

other quantizations Models above have a default quantization of Q4_K_M. To run other quantizations (e.g., Q8): ollama run ibm/granite4:tiny-h-q8_0

base models Base models without instruction tuning are provided for all sizes and quantizations. These can be accessed with tags such as ibm/granite4:tiny-h-base-f16.

Supported Languages

Supported Languages: English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese. Users may finetune Granite 4.0 models for languages beyond these languages.

Intended Use

This model is designed to handle general instruction-following tasks and can be integrated into AI assistants across various domains, including business applications.

Capabilities

• Summarization
• Text classification
• Text extraction
• Question-answering
• Retrieval Augmented Generation (RAG)
• Code related tasks
• Function-calling tasks
• Multilingual dialog use cases
• Fill-In-the-Middle (FIM) code completions

Learn more

• Developers: Granite Team, IBM
• Website: Granite Docs
• GitHub Repository: ibm-granite/granite-4.0-language-models
• Release Date: October 2nd, 2025
• License: Apache 2.0