Granite 4.2 8B

Granite 4.2 8B is a dense, decoder-only reasoning language model developed by IBM Research and released as part of the Granite 4.2 family. It is positioned as a mid-sized, balanced model designed for general-purpose enterprise applications, offering a compromise between the efficiency of the 3B variant and the performance of the 30B flagship. Built on a transformer architecture utilizing Grouped Query Attention (GQA) and Rotary Position Embeddings (RoPE), the model is pre-trained on approximately 15 trillion tokens before undergoing a rigorous post-training process.

Historically, the Granite family focused on instruction-following efficiency, but the 4.2 generation introduces native chain-of-thought (CoT) reasoning capabilities. The model uses explicit <think>...</think> tags to deliberate before generating a final response, which significantly improves performance on complex multi-step logic, mathematical reasoning, and coding tasks. Users can toggle between three distinct thinking modes: a high-reasoning default, a non-thinking mode for lower latency, and a low-effort mode that adapts deliberation depth to the complexity of the query.

A key distinction for the 8B variant is its inclusion in the agentic reinforcement learning block of the training pipeline. Unlike the smaller 3B model, the 8B version was trained inside live sandboxed environments—including terminal shells and web browsers—to perform real software engineering tasks and multi-hop search queries. This training enables more reliable reasoning-augmented tool calling, where the model evaluates which tools to invoke and in what order before executing actions.

The model supports a native 128,000-token context window, with documentation indicating potential extensions up to 512,000 tokens for handling long-form documents. It is multilingual by design, with validated performance across 12 languages including English, German, French, Spanish, Japanese, and Chinese. Distributed under the Apache 2.0 license, Granite 4.2 8B is intended for open commercial and research use, emphasizing transparency with published training recipes and ISO-certified data governance standards.