Gemma 4 31B IT is the largest dense variant in the fourth-generation Gemma family of open-weights models from Google DeepMind, published under an Apache 2.0 license. The family is described as built from Gemini 3 research with a focus on maximizing intelligence per parameter, and it ships in five sizes (E2B, E4B, 12B, 26B A4B, and 31B) spanning both dense and mixture-of-experts architectures. The instruction-tuned 31B version is positioned for text generation, coding, and reasoning workloads, with configurable thinking modes that let applications tune how much deliberation the model performs before answering.

The 31B IT model is multimodal on the input side, accepting both text and images, and produces text output, while audio input is limited to the smaller E2B, E4B, and 12B variants in the same family. It is designed to operate over a very large context window of up to 256K tokens, making it well suited to long-document analysis, codebase reasoning, and multi-turn agentic flows. Multilingual coverage extends to over 140 languages, which broadens its practical fit for global applications. Through UnoRouter's unified API, this variant is exposed as a free-tier route, giving developers a no-cost path to experiment with one of the most capable open models in the Gemma 4 lineup.