IBM has introduced Granite 4.2, the latest addition to its family of open-weight large language models designed for download and self-hosting. The lineup includes 3B, 8B, and 30B parameter variants, all decoder-only with a native 128,000-token context window.
The 8B and 30B versions also underwent agentic reinforcement learning to strengthen capabilities such as terminal use, web search, and external tool interaction. The 3B model can use tools as well, though without the same specialized training.
Comments