Granite 4.0 Micro

Granite 4.0 Micro is a 3-billion parameter large language model developed by IBM as part of the Granite 4.0 family of enterprise-focused models. Released in October 2025, it utilizes a conventional decoder-only dense transformer architecture, serving as an alternative to the hybrid Mamba/transformer models in the same series for environments where hybrid architectures are not yet optimized.

The model was pre-trained on a corpus of approximately 15 trillion tokens across four training stages, incorporating a mix of open-source and proprietary data with a focus on code, mathematics, and high-quality synthetic content. It is designed to balance strong performance with low latency and memory requirements, making it suitable for high-volume, low-complexity tasks such as text extraction, summarization, and Retrieval Augmented Generation (RAG).

Key capabilities include advanced tool-calling and instruction-following, optimized for building agentic workflows and AI assistants. It supports long-context processing and is natively multilingual, covering twelve languages including English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese. The model is released under the Apache 2.0 license.