IBM Unveils Granite 4.2: Reasoning Language Models That Think and Act Inside Environments

IBM has released Granite 4.2, its first family of dense, decoder-only reasoning language models, in three sizes: 3B, 8B, and 30B parameters.

The release, announced on August 25, 2026, gives every Granite model a switchable thinking mode and sends the two larger sizes through reinforcement learning inside real software-engineering, terminal, and web-search environments.

Granite 4.2 models are trained from scratch on roughly 15 trillion tokens, with a five-phase schedule that extends the context window to 512K tokens.

The release's center of gravity is its multi-stage reinforcement learning pipeline, where each stage trains one capability in a separate training run, warm-started from the previous stage's checkpoint.