Google has launched Gemma 4, a new open-source AI model, under the Apache 2.0 license. This model allows users to run applications locally on devices without needing an internet connection, affirming digital sovereignty and developer flexibility.
Gemma 4 can operate on billions of Android devices and select laptop GPUs. It represents a significant evolution from its predecessor and is positioned as Google’s most capable open AI model to date, distinguishing itself from the proprietary Gemini AI product.
The model offers advanced reasoning capabilities, including multi-step planning and enhanced logic. Google stated that it has made significant advancements in math and instruction-following benchmarks with this iteration.
Gemma 4 supports audio and video processing for speech recognition and visual interpretation. It is available in four different sizes—two billion, four billion, 26 billion, and 31 billion weights—providing options for developers.
Previous versions of Gemma were open-weight but not open-source, meaning users could not freely modify or distribute them. The Apache 2.0 license for Gemma 4 permits users to download, modify, and use the model for personal or commercial purposes, provided they give proper attribution.
Gemma 4 can be accessed through Google AI studio and is also available for download from third-party platforms such as Hugging Face, Kaggle, or Ollama. This transition to open-source is expected to facilitate broader application integration without recurring subscription fees.
According to a Google blog post, “This open-source license provides a foundation for complete developer flexibility and digital sovereignty; granting you complete control over your data, infrastructure, and models.” This development places Gemma 4 in a unique position within the AI landscape, appealing to developers seeking more control and privacy in their AI applications.
