Granite 4.2 LLMs: How They're Built
Granite 4.2 is IBM and the Granite Team’s first family of dense, decoder-only reasoning LLMs, released in three sizes (3B, 8B, 30B) and trained from scratch on ~15T tokens with a pipeline that includes long-context extension to 512K tokens, supervised fine-tuning on chain-of-thought/reasoning/agentic-trajectory data, and multi-stage reinforcement learning with agentic RL in sandboxed environments.
Aug 25, 2026 ·Hugging Face


















