ibm-granite/granite-4.1-8b · Hugging Face

IBM's Granite team released Granite-4.1-8B, an 8 billion parameter long-context instruct model on Hugging Face under the permissive Apache 2.0 license. This model is fine-tuned from Granite-4.1-8B-Base using a mix of open-source instruction datasets and internally collected synthetic data. It features an improved post-training pipeline combining supervised fine-tuning and reinforcement learning alignment, which significantly enhances tool calling, instruction following, and chat capabilities. The model supports 12 languages: English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese, with users able to fine-tune it for additional languages.

The intended use cases are broad, including general instruction following, AI assistants across business domains, and LLM agents with tool-use capabilities. Specific capabilities include summarization, text classification, text extraction, question-answering, Retrieval Augmented Generation (RAG), code-related tasks, function-calling, multilingual dialog, and Fill-In-the-Middle (FIM) code completions. The model is available through IBM's Hugging Face collection, with a technical blog and GitHub repository providing further details. Released on April 29, 2026, Granite-4.1-8B represents IBM's commitment to open, enterprise-ready AI with strong multilingual and tool-use support.