IBM announced Granite 4.0, its next-generation family of open-source language models featuring a hybrid Mamba/transformer architecture that cuts memory use and costs while maintaining high performance. Available in multiple sizes (Micro, Tiny, Small), these models are optimized for enterprise workflows, agentic AI tasks, and edge applications, offering over 70% RAM savings and faster inference compared to traditional LLMs. Granite 4.0 has received positive press coverage, ISO 42001 certification for safe and responsible AI, and early testing by enterprise partners like EY and Lockheed Martin.
