Gemma 4 31B IT is an instruction-tuned member of the Gemma 4 family of open models created by Google DeepMind, positioned as the most capable variant among five released sizes (E2B, E4B, 12B, 26B A4B, and 31B). The family is described as built from Gemini 3 research and technology, with a focus on maximizing intelligence per parameter. Across the lineup, Gemma 4 combines Dense and Mixture-of-Experts architectures, giving the 31B IT model a lineage that draws on both efficient mobile-oriented designs and higher-capacity server-class deployments. It is distributed under the Apache 2.0 license as open weights, making it well suited for teams that want to self-host, fine-tune, or audit a state-of-the-art model without proprietary restrictions.
The model is designed for practical text generation, coding, and reasoning workloads, with configurable thinking modes that let users dial up or down the depth of deliberation. It accepts text and image input while producing text output, sitting within a context window of up to 256K tokens and maintaining multilingual coverage in over 140 languages. Its combination of a long context, strong reasoning orientation, and multimodal input makes it a flexible choice for assistants, document understanding, code generation, and complex multi-step analysis, particularly in environments where open weights and self-hosting are priorities.
