Granite 4.0 Micro

Granite 4.0 Micro is IBM's 3B-parameter instruct model in the Granite 4.0 family, designed for long-context work and enterprise integration. It is instruction-tuned from Granite-4.0-Micro-Base using a blend of permissive open-source instruction datasets and internally generated synthetic data, then refined with supervised finetuning, reinforcement-learning alignment, and model merging. The Granite 4.0 instruct line emphasizes stronger instruction following and tool-calling, positioning the Micro variant as a practical small-model option for structured, tool-driven workflows. Independent third-party listings and tutorials place the Micro checkpoint alongside the Nano, Tiny, and Small siblings, with the family as a whole reported to use IBM's Hybrid H Mamba architecture trained on 15 trillion tokens, a design aimed at faster inference and lower memory use at small scales.

In day-to-day use, the model's compact 3B size makes it attractive for local or cost-sensitive deployments where it can still handle longer prompts and call external tools through its supported tool-calling capability. Its post-training recipe, combining RL alignment and model merging on top of broad open-source and synthetic data, supports reliable handling of structured chat formats and step-by-step instructions. Developers working on support agents, automation pipelines, or retrieval-augmented assistants may find the Micro variant a balanced choice when they need enterprise-oriented instruction following without the resource demands of larger hybrid models, while still benefiting from the architectural efficiency improvements IBM introduced across the Granite 4.0 line.