IBM Releases Granite 4.2, Open Reasoning Language Models with Agentic Capabilities

IBM has released Granite 4.2, an open reasoning language model family in three sizes: 3B, 8B, and 30B parameters. Unlike earlier versions of Granite, which were instruction-following assistants, Granite 4.2 is built around explicit reasoning.

The models can emit a chain of thought before answering and expose a thinking/non-thinking switch, as well as a low-effort mode that spends a short reasoning budget on easy questions.

Granite 4.2 uses decoder-only dense transformers and is pre-trained from scratch on roughly 15 trillion tokens. The 8B and 30B models include an agentic RL block where the model learns to edit code, drive a terminal, and run web searches inside real sandboxed environments.

The models are available under Apache 2.0 and can be used for fine-tuning and commercial production. IBM also released two 470M-parameter Granite Speech 5.0 Turbo CTC models alongside the LLMs.