← Back to the wire

Granite 4.2 LLMs: How They're Built

AnnouncementModelAug 25, 2026

IBM released Granite 4.2, a family of dense, decoder-only reasoning LLMs in 3B, 8B, and 30B sizes, all under the Apache 2.0 license. Each model is pre-trained from scratch on roughly 15 trillion tokens, with a five-phase strategy extending the context window to 512K tokens, then fine-tuned and post-trained with multi-stage reinforcement learning. The 8B and 30B models undergo agentic RL in sandboxed environments, and every model includes a thinking/non-thinking switch and native tool calling.

Receipt № 15621 source · awaiting confirmation ◐

Evidence

1source· awaiting independent confirmation

No score is assigned. Sources and their independence are shown in the citation chain below.

Citation chain · 1 source

IBMCompanyGranite 4.2Model
Canonical: https://huggingface.co/blog/ibm-granite/granite-4-2