IBM's new Granite 4.2 models ride the wave of interest in local LLMs

Ars Technica reports that IBM has introduced the Granite 4.2 series, a set of open‑weight large language models that users can download and run on their own hardware. The lineup includes three‑billion, eight‑billion and thirty‑billion parameter versions, all built with a decoder‑only architecture and a native context window of one hundred twenty‑eight thousand tokens. The larger eight‑billion and thirty‑billion models incorporate an agentic reinforcement‑learning stage that equips them to operate a terminal, browse the web, or invoke external tools, while the smallest model also supports tool use but without that specialized training. These releases extend IBM's effort to provide locally hosted AI alternatives with expanded tool‑use capabilities.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton