IBM has unveiled Granite 4.2, the latest generation of its open-source language model family, designed specifically with enterprise deployment in mind. According to MarkTechPost, the release includes models in three sizes—3B, 8B, and 30B parameters—all distributed under the permissive Apache 2.0 license, giving organizations full freedom to adapt and deploy them commercially.

What sets Granite 4.2 apart is its built-in flexibility around reasoning depth. Each model includes a switch that lets users choose between thinking, low-effort, and non-thinking modes, allowing teams to balance response quality against computational cost depending on the task at hand. The models also come with native tool-calling capabilities, making them well suited for agentic workflows where AI systems need to interact with external systems or software.

Notably, the 8B and 30B versions have undergone additional agentic reinforcement learning, training them specifically to identify and edit code—a capability with real implications for software development and IT automation teams. By open-sourcing this level of enterprise-grade reasoning capability, IBM is lowering the barrier for businesses to build customized, cost-effective AI solutions without relying solely on closed, proprietary systems.

Releases like this reflect a broader shift toward practical, enterprise-ready open models. At Generative AI Solutions, we track these developments closely to help clients choose the right foundation for their AI initiatives.