Gemma 4 31B IT

Gemma is a family of open models built by Google DeepMind. The Gemma 4 models are multimodal models that support processing text and image inputs (E2B, E4B, and 12B also support audio) and generating text outputs. This release includes open-weight models in both pretrained and instruction-tuned variants. Gemma 4 supports multilingual capabilities across more than 140 languages. Gemma 4 includes dense and Mixture-of-Experts (MoE) architectures, suitable for tasks such as text generation, programming, and reasoning. The model series is available in five different sizes: E2B, E4B, 12B, 26B A4B, and 31B, and the diverse sizes allow it to be deployed in environments ranging from high-end mobile phones to laptops and servers.