Source
Granite 4.2 LLMs: How They're Built
IBM has released Granite 4.2, a family of dense, decoder-only reasoning LLMs available in 3B, 8B, and 30B sizes. These models emphasize explicit reasoning and can produce a chain of thought before providing final answers.
Why it matters
These models enable developers to create AI agents capable of editing code, using terminals, and searching the web. The Apache 2.0 license allows for open and flexible integration into various software environments.
The details
- Models are pre-trained on 15 trillion tokens with a 512K token context window.
- The 8B and 30B models feature agentic RL for real-world tool operation.
- Capabilities include a thinking/non-thinking switch and native OpenAI-compatible tool calling.
Get the weekly recap
The stories like this one, picked and explained — once a week, straight to your inbox.