IBM Granite 4.2 targets enterprise agent reasoning with low-cost deployment
Granite 4.2 focuses on enterprise AI agents rather than consumer chatbots, aiming for reasoning that is cheaper and dependable at scale. IBM describes an 8B model for coding and tool use on a single GPU and a 30B flagship for multi-step reasoning. It offers reasoning modes to balance speed and cost, trained for tool and terminal tasks, with IBM Bob and Consulting already using it for development automation. The plan is private, on-prem deployment.