IBM has launched Granite 4.2, the latest set of open-weight large language models designed for local deployment in enterprise environments. The new family includes 3B, 8B, and 30B parameter variants, all built with a decoder-only architecture.
The 8B and 30B models incorporate an agentic reinforcement learning block, enabling expanded tool-use capabilities such as terminal interaction, web searches, and external tool integration. The 3B variant supports tools but lacks the specialized training of its larger counterparts. All models feature a native 128,000-token context window, doubling the capacity of earlier releases.
The update targets predictable enterprise deployment, emphasizing agentic behavior and self-hosted operation. IBM positions these models as part of its strategy to meet growing demand for locally run LLMs in business settings.
