Technology
IBM releases Granite 4.2 open-weight language models for local deployment
The new open-weight models feature up to 30 billion parameters and specialized training for external tool use.
The short version
- IBM released its Granite 4.2 open-weight large language model family, designed for self-hosting and local enterprise deployment.
- The models come in 3B, 8B, and 30B parameter variants and feature a native 128,000-token context window.
- The larger variants include specialized reinforcement learning for agentic tasks like terminal operations and web searches.
Key facts
- IBM launched the Granite 4.2 open-weight large language model family designed for local downloading and self-hosting.[Ars Technica]
- The decoder-only Granite 4.2 models are available in 3B, 8B, and 30B parameter variants.[Ars Technica]
- All Granite 4.2 model variants feature a native 128,000-token context window.[Ars Technica]
- The 8B and 30B parameter variants underwent specialized agentic reinforcement learning training to handle tasks such as searching the web, executing terminal commands, and using external tools.[Ars Technica]