IBM releases Granite 4.1 8B, an Apache 2.0 instruct model for enterprise tasks

IBM released Granite-4.1-8B, an 8B-parameter long-context instruct model fine-tuned from Granite-4.1-8B-Base. The card gives a release date of April 29th, 2026 and credits the Granite Team at IBM. It says Granite 4.1 went through an improved post-training pipeline, including supervised fine-tuning and reinforcement learning alignment, that improves tool calling, instruction following and chat.

The training mix is open instruction datasets with permissive licences and internally collected synthetic data. The listed languages are English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch and Chinese. The intended uses are summarisation, classification, extraction, question answering, retrieval-augmented generation and tool-using agents.

The licence is Apache 2.0. A Hugging Face collection, technical blog and GitHub repository are linked from the card.

Not a headline model. It is the kind a legal department can say yes to.

Sources

• Model card on Hugging Face
• Hugging Face repo metadata (creation date, licence, size)

Editor's first pass (AI-assisted)

• Real 4.5: Apache 2.0 and ungated, with a linked collection, blog and repository.
• Useful 3.5: Summarisation, extraction, RAG and tool use for business assistants.
• Notable 3: 256 likes; steady rather than leading.
• Novel 2.5: Improved post-training on the Granite 4 line, per the card.