Granite 4.2 LLMs: How They're Built
huggingface ·
Granite 4.2 is IBM's first family of dense, decoder-only reasoning LLMs, released in 3B, 8B, and 30B sizes. The models are pretrained from scratch on about 15 trillion tokens using a five-phase strategy that extends the context window to 512K tokens. They are then supervised fine-tuned on chain-of-thought, reasoning, and agentic-trajectory data, followed by a multi-stage reinforcement learning pipeline. The 8B and 30B models also undergo agentic RL inside real sandboxed environments to learn tool use, code execution, terminal interaction, and web search. Every model supports a thinking/non-thinking switch, a low-effort thinking mode, and native tool calling in an OpenAI-compatible format. The training pipeline uses asynchronous GRPO and includes data quality control with LLM-based judges and deduplication. All Granite 4.2 models are released under the Apache 2.0 license.
huggingface