IBM Unveils Granite 4.2 Language Models with Built-in Agentic Capabilities

IBM has released its Granite 4.2 language models in three sizes: 3B, 8B, and 30B. These models were trained from scratch on approximately 15 trillion tokens and support context windows up to 512,000 tokens. According to IBM, the models can switch between 'thinking' and 'non-thinking' modes to control compute per task, with a 'low-effort' mode that saves resources for simple queries.

The 8B and 30B variants also underwent agentic RL training, where they learned to use tools, write and run code, and search the web in real sandbox environments. This allows them to perform tasks such as tool calling, which is supported by all models in OpenAI format. They can run on vLLM or SGLang.

Benchmark results for the three Granite 4.2 models on agentic tasks and tool use show that the 30B model leads across all three tests.