Granite 4.2 LLMs: How They're Built

Granite 4.2 is the first family of dense, decoder-only reasoning LLMs from IBM, released in three sizes: 3B, 8B, and 30B. Each model is pre-trained on approximately 15 trillion tokens and features a five-phase training strategy, including supervised fine-tuning and a multi-stage reinforcement learning pipeline. The models support native tool calling and have a thinking/non-thinking switch for different reasoning modes.