IBM's Granite 4.2 models differ in more than size. All three support tool calls, but the 8-billion- and 30-billion-parameter versions receive additional training through tasks in working software environments. The 3-billion-parameter model does not receive that training block. IBM's technical walkthrough explains the distinction.
For software-engineering tasks, the larger models work in a sandboxAn isolated environment where AI-generated code or agent actions run without being able to touch anything real.Full definition → repository, and hidden tests determine the reward. Terminal tasks provide feedback on successful completion, while search tasks use a model to judge the final answer. This trains sequences of actions against their outcomes, giving the larger models experience beyond formatting a valid tool call.
Terms in this piece · Glossary
sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
chain-of-thought — Having a model write out intermediate reasoning steps before its answer, which markedly improves performance on hard problems.
tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
