Granite 4.2 Brings Native Reasoning to Enterprise Agents

IBM's Granite language and speech models have been upgraded to version 4.2, which includes native reasoning capabilities for enterprise agents.

The new model is designed to complete tasks rather than just generate responses, and it can plan before acting, weigh trade-offs, and catch mistakes. It's built for complex workflows in enterprise settings, where tasks often involve multiple steps and ambiguous instructions.

Granite 4.2 features a dense architecture that supports broad compatibility across cloud, on-premises, and edge environments. The model is available in three sizes: 3B, 8B, and 30B parameter sizes.