Google Releases Multimodal Gemma 4 31B Model

Google DeepMind has expanded its open model lineup with the release of Gemma 4 31B Instruct. This new model is a dense, 31-billion-parameter foundation model, marking a major update and the first entry in the Gemma 4 series.

The most significant feature of this release is its native multimodality. Described as an "any-to-any" model, Gemma 4 moves beyond text-only interactions to incorporate vision capabilities, allowing it to process and understand both text and images. This positions the model as a powerful tool for building more sophisticated, context-aware AI applications.

As an instruction-tuned model—specifically designated as an "assistant" variant—Gemma 4 31B has been fine-tuned to follow user prompts and engage in helpful dialogue. This makes it well-suited for direct integration into chatbots, creative tools, and other interactive products.

Continuing its commitment to open development, Google has released Gemma 4 under the commercially-friendly Apache 2.0 license. This allows developers and businesses to freely use, modify, and deploy the model. The official weights and model card are available on Hugging Face.