Granite 4.2 8B

Granite 4.2 8B is IBM's small dense reasoning model, and it suits people who need a permissively licensed model that fits modest hardware. The Hugging Face card gives an Apache 2.0 licence and a release date of 25 August 2026, with a native 128K context that can be extended to 512K. It was tested in 12 languages, and the card says other languages may work but are not fully tested. The card lists 8B parameters, while its file metadata shows 9B, so the two do not match. OpenRouter lists it from 31 August 2026 with a 131,072-token window, text only, at $0.06 per million input tokens and $0.25 per million output tokens. The card warns that it can produce inaccurate or unsafe answers, that thinking output may be unpolished, and suggests IBM's Granite Guardian alongside it. Figures come from the Hugging Face card and the OpenRouter model list. We have not run it ourselves. We would use it for cheap classification and light coding help, not for hard agent work.